Claims:CLAIMS
We claim:
1. An apparatus for biometric identification of an animal, the apparatus comprising:
a trevis assembly providing an enclosure around the animal;
a head rest affixed to the trevis assembly and adapted to support a head of the animal in a substantially fixed position;
a lighting control assembly affixed to the trevis, the lighting control assembly comprising:
a soft light source adapted to illuminate a muzzle of the animal with soft, diffused and glare-free light; and
a light diffuser adapted to prevent the glare of strong ambient light sources on the muzzle; and
a user device mount adapted to position a user device substantially in front of the muzzle.
2. The apparatus according to claim 1, wherein the trevis assembly is adapted to be disassembled.
3. The apparatus according to claim 1, wherein the head rest is adapted to be adjusted in height in accordance with a height of the animal.
4. The apparatus according to claim 1, wherein the user device mount is adapted to allow adjustment of the position the user device at a desired distance and a desired orientation with respect to the muzzle.
5. The apparatus according to claim 1 further comprising a user device.
6. The apparatus according to claim 5, wherein the user device comprises an image capture module configured to facilitate capturing a muzzle image suitable for use in biometric identification.
7. The apparatus according to claim 5, wherein the user device comprises a user identity module configured to provide an identity of a user of the user device.
8. The apparatus according to claim 5, wherein the user device comprises a location module configured to detect a geographical location of the user device.
9. The apparatus according to claim 5 further comprising an animal database storing a plurality of animal records, wherein each animal record pertains to a registered animal and comprises muzzle print information of the registered animal.
10. The apparatus according to claim 5 further comprising an identification server.
11. The apparatus according to claim 1 further comprising a feature extractor configured to compute a feature representation of a muzzle image.
12. The apparatus according to claim 9 further comprising a classifier configured to compare a feature representation of a muzzle image to muzzle print information of registered animals.
13. The apparatus according to claim 9 further comprising includes a shortlisting module to shortlist the registered animals to be considered during the identification process based on one or more shortlisting parameters.
14. The apparatus according to claim 12, wherein the shortlisting module shortlists registered animals based at least one of a location of the registered animal and an owner of the registered animal.
Description:

FIELD OF THE INVENTION
The present invention generally relates to identification of animals, and more particularly to biometric identification of animals based on their muzzle prints.
PRIOR ART AND PROBLEM TO BE SOLVED
Animal identification, and particularly cattle (buffalo, horse, and cow) identification, is important for various applications such as animal healthcare, understanding disease trajectory, vaccination and production management, animal traceability, and animal ownership assignment.
Traditionally, solutions such as ear tags, branding, tattooing, and the like have been used to identify animals. More recently, electronic methods such as Radio Frequency Identification (RFID) tags, injectable tags, barcodes, Quick Response (QR) codes, and the like have also been used. However, the performance of these methods is limited due to their vulnerability to losses, duplications, fraud, harmful implanting, complicated retrieval system and security challenges. Further, typically such methods involve discomfort to the animals due to tags, codes, and the like, being added to the animal’s body.
Many animals, such as cows, buffalos, horses, sheep, goats, dogs, cats, and so on, can be uniquely identified by their muzzle prints. The arrangement of beads and ridges on the muzzle of these animals is unique to each animal, in much the same way as fingerprints are unique to each human being. Thus, the muzzle print of such animals can be used for biometric identification. Biometric identification offers many advantages. It is unique, immutable, inexpensive, and does not require any uncomfortable or painful procedures for the animal. However, it also introduces some challenges, such as proper image acquisition, identification accuracy, processing time, and overall system operability.
It is challenging to acquire images suitable for biometric identification in real-world applications. On one hand, the image processing algorithms used in biometric identification tend to be sensitive to image quality. Poor quality input images can lead to failure in identification and/or incorrect identification. On the other hand, field conditions are often not conducive for conveniently capturing muzzle images suitable for biometric identification. Environmental factors such as lighting, movement of the animal, the distance between the animal’s muzzle and the camera, the angle from which the image is acquired, the presence of moisture, feed, dirt, etc. on the muzzle, and so on, all impact the quality of the acquired image, and may render it unsuitable for use in biometric identification.
There is a need for robust solutions for biometric identification of animals that mitigate the effect of environmental factors and provide convenient, fast, and accurate identification in the field.
OBJECTS OF INVENTION
It is an object of the present invention to provide an improved solution for muzzle print based biometric identification of animals.
It is another object of the present invention to facilitate capture of muzzle images that are suitable for use in biometric identification.
It is another object of the present invention to improve the accuracy of muzzle print based biometric identification of animals.
It is another object of the present invention to provide a solution for biometric identification of animals in cow shelters or ‘gaushalas’ or cattle owners.
It is another object of the present invention to provide a solution for biometric identification of animals in an accurate, cost-efficient, convenient, and scalable manner by following an innocuous process.
It is still another object of the present invention to provide a portable solution for biometric identification of animals that can be deployed anywhere with minimal infrastructure pre-requisites.
SUMMARY OF INVENTION
In various embodiments, the present invention discloses an apparatus for biometric identification of an animal. The apparatus includes a trevis assembly providing an enclosure around the animal, a head rest affixed to the trevis assembly and adapted to support a head of the animal in a substantially fixed position, a lighting control assembly affixed to the trevis, and a user device mount adapted to position a user device substantially in front of the muzzle. The lighting control assembly includes a soft light source adapted to illuminate a muzzle of the animal with soft, diffused and glare-free light, and further includes a light diffuser adapted to prevent the glare of strong ambient light sources on the muzzle.
In an embodiment, the trevis assembly is adapted to be disassembled, and head rest is adapted to be adjusted in height in accordance with a height of the animal. Further, the user device mount is adapted to allow adjustment of the position the user device at a desired distance and a desired orientation with respect to the muzzle.
In another embodiment, the apparatus further includes a user device. The user device includes an image capture module configured to facilitate capturing a muzzle image suitable for use in biometric identification, a user identity module configured to provide an identity of a user of the user device, and a location module configured to detect a geographical location of the user device.
In an embodiment, the apparatus includes an animal database storing a plurality of animal records, wherein each animal record pertains to a registered animal and comprises muzzle print information of the registered animal. In an embodiment, the apparatus includes an identification server.
In an embodiment, the apparatus includes a feature extractor configured to compute a feature representation of a muzzle image, and a classifier configured to compare a feature representation of a muzzle image to muzzle print information of registered animals.
BRIEF DESCRIPTION OF DRAWINGS
FIG. 1 shows an overview of a smart muzzle identification (ID) system according to an embodiment of the present invention.
FIGs. 2A and 2B illustrate an apparatus for muzzle image capture for biometric identification of animals according to an embodiment of the present invention.
FIGs. 3A-3D illustrate a trevis assembly according to an embodiment of the present invention.
FIG. 4 shows example training images used to train a preliminary classifier according to an embodiment of the present invention.
FIG. 5 shows a block diagram illustrating the training of a preliminary classifier for rejecting non-muzzle images according to an embodiment of the present invention.
FIG. 6 shows a block diagram illustrating registration of a new animal with the smart muzzle ID system (SMIS).
FIG. 7 shows example images captured at the time of registration of an animal with SMIS according to an embodiment of the present invention.
FIG. 8 shows a block diagram illustrating identification of an animal with the SMIS.
FIG. 9 shows an example user interface for guiding the user during capture of a muzzle image according to an embodiment of the present invention.
FIG. 10 shows example images at various steps during the muzzle image capture process according to an embodiment of the present invention.
FIG. 11 shows screen representations illustrating an interface for secondary validation according to an embodiment of the present invention.
FIG. 12 shows the distribution of responsibilities between a user device and an identification server according to an embodiment of the present invention.
FIG. 13 shows the distribution of responsibilities between a user device and an identification server according to another embodiment of the present invention.
FIG. 14 shows an exemplary user device in accordance with an embodiment of the invention.
DETAILED DESCRIPTION OF INVENTION
The following is a detailed description of example embodiments to illustrate the principles of the invention. The embodiments are provided to illustrate aspects of the invention, but the invention is not limited to any embodiment. The scope of the invention encompasses numerous alternatives, modifications and equivalent; it is limited only by the claims.
Further, throughout this disclosure, the singular terms “a,” “an,” and “the” include plural referents unless the context clearly indicates otherwise. Similarly, the word “or” is intended to include “and” unless the context clearly indicates otherwise. The word “animal” is intended to include the broad genus of animals, including the subgenera of ruminant animals and animals raised for food production, unless the context clearly indicates otherwise.
Numerous specific details are set forth in the following description in order to provide a thorough understanding of the invention. However, the invention can be practiced according to the claims without some or all of these specific details. For the purpose of clarity, technical material that is known in the technical fields related to the invention has not been described in detail so that the invention is not unnecessarily obscured.
The present invention provides solutions for reliable, accurate, and convenient biometric identification of animals based on their muzzle prints. In various embodiments, the present invention uses robust identification algorithms, including state of the art machine learning and artificial intelligence technology, to achieve improved identification performance. Further, the present invention employs a combination of physical and algorithmic optimizations to reduce and/or compensate for the effect of environmental factors during muzzle image capture. This allows capturing of good quality muzzle images to be used by the identification algorithms, thereby further improving identification performance. In various embodiments, the teachings of the present invention reduce the time, computation, and expense needed for animal identification.
OVERVIEW
FIG. 1 shows an overview of a smart muzzle identification (ID) system according to an embodiment of the present invention. A smart muzzle ID system (SMIS) 102 identifies an animal 104 based on the unique muzzle print of animal 104. SMIS 102 includes a user device (UD) 106, an identification server (IDS) 108, and an animal database (AD) 110.
SMIS 102 conducts an animal registration process (interchangeably referred to herein as the “training phase”, “training step”, “registration step”, “registration phase”, “registration process”, or simply “registration”) in which animal 104 is registered on SMIS 102 and relevant information for identifying animal 104 (hereinafter collectively referred to as the “animal identity profile”) is obtained and stored in AD 110. In various embodiments, a user (not shown) of UD 106 captures one or more muzzle images of animal 104 at the time of its registration using a camera in UD 106. Various provisions of the present invention facilitate capture of muzzle images suitable for use in biometric identification, as discussed later in this description. Said muzzle images and/or their derived representations (jointly and severally referred to herein as “muzzle print information”) form a part of the animal identity profile and are used to train machine learning algorithms embodied in SMIS 102 to recognize animal 104. All animals to be identified are similarly registered on SMIS 102 (hereinafter collectively referred to as the “registered animals”). The animal identity profiles of the registered animals, including their muzzle print information, are stored in AD 110.
Once trained, SMIS 102 can test a muzzle image to determine if it is of one of the registered animals or not. This process is interchangeably referred to herein as the “testing phase”, “testing step”, “identification step”, “identification phase”, “identification process”, or simply “testing” or “identification” in this description.
In order to identify animal 104, the user operates UD 106 to capture a new muzzle image of animal 104 that is suitable for use in muzzle-based biometric identification of animal 104. In an embodiment, SMIS 102 includes a feature extractor (not shown; hereinafter “FE”) which computes a feature representation of the new muzzle image. Further, in an embodiment, SMIS 102 includes an identification classifier (not shown) that uses said feature representation of the new muzzle image to classify it as corresponding to one (or none) of the registered animals using at least the muzzle print information contained in the animal identity profiles of the registered animals. In other words, the identification classifier determines which of the registered animals has muzzle print information that most closely resembles the new muzzle image. In an embodiment, if the new muzzle image is found not to resemble any of the registered animals, SMIS 102 initiates the registration process for animal 104.
UD 106 and IDS 108 are communicatively coupled through a network 112. In various embodiments, UD 106 and IDS 108 exchange information such as muzzle images and feature representations with each other over network 112. Further, in various embodiments, UD 106 and IDS 108 exchange commands over network 112, for example for allowing UD 106 to access and/or update AD 110. In various embodiments, information stored in AD 110 is accessed and/or updated by at least one of IDS 108 and UD 106.
In various embodiments, the feature extractor is embodied in either UD 106 or IDS 108. Similarly, in various embodiments, the identification classifier is embodied in either UD 106 or IDS 108. While various exemplary embodiments are illustrated in this description, it will be apparent to a person skilled in the art that the constituent functionalities of SMIS 102 can be embodied in one or more of UD 106, IDS 108, and AD 110, without deviating from spirit and scope of the present invention.
TREVIS ASSEMBLY
FIGs. 2A and 2B illustrate an apparatus for muzzle image capture for biometric identification of animals according to an embodiment of the present invention. FIG. 2A shows a perspective view of the apparatus, while FIG. 2B shows an exploded view.
FIG. 2A shows animal 104 (only head shown) placed in front of UD 106 in a position suitable for capture of a muzzle image. Animal 104 is enclosed by a trevis assembly 202, which inter alia serves to reduce motion of animal 104 and prevent motion blur in the muzzle image. Further, affixed to trevis assembly 202 is a head rest 204 for resting the head of animal 104 in a manner that further restricts movement, thereby substantially eliminating motion blur from the captured muzzle image. In an embodiment, the position and dimensions of head rest 204 can be adjusted to comfortably accommodate animals of various different sizes. Various mechanisms for facilitating said adjustments are known or may be developed in the future, and any such suitable mechanism can be used in conjunction with the present invention without deviating from its spirit and scope.
Further, the figure shows UD 106 mounted on a user device mount (UDM) 206. UDM 206 significantly reduces, or eliminates, the motion blur caused due to any movement of UD 106 (which may ordinarily be expected if UD 106 is held in the user’s hand without any external support). Further, UDM 206 allows adjustment of the height, position, and orientation of UD 106. This allows the user to position UD 106 optimally for capturing a good quality muzzle image of animal 104.
In various embodiments, trevis assembly 202, head rest 204, and UDM 206 are so configured that the distance between animal 104’s muzzle and UD 106 is between six (6) and ten (10) inches (6”-10”), and UD 106 is at substantially the same height as, and directly in front of, animal 104’s muzzle.
The distance of 6”-10” is far enough to allow clear and sharp focus by most conventional cameras, while being close enough to obtain a relatively high-resolution capture of the muzzle area of large animals such as cows. It will be apparent to a person skilled in the art that for other animals, the desired distance between the muzzle and UD 106 may be suitably adjusted up or down depending on the dimensions of the animal’s muzzle and/or the specifications of the camera of UD 106.
Further, placing UD 106 at substantially the same height as, and directly in front of, animal 104’s muzzle ensures that the muzzle image captures substantially all of the muzzle, and further that it captures it from a substantially consistent angle and distance (since training and testing images of an animal are captured using similar equipment and procedure).
The foregoing configuration thus minimizes the scope for distance and angle variations during image capture that may negatively affect the performance of SMIS 102.
Further, a lighting control assembly (LCA) 208 is shown connected to trevis assembly 202 and disposed slightly above and ahead of animal 104’s head. LCA 208 provides soft lighting for animal 104’s muzzle so that it is well-lit and free from glare. In the illustrated embodiment, LCA 208 includes an umbrella light diffuser (ULD) 210 and a soft light source (not visible) disposed within ULD 210. ULD 210 prevents the glare of strong ambient light sources on the muzzle of the animal. In an embodiment, ULD 210 protects the muzzle from strong ambient light sources that may cause a glare on the muzzle, such as the sun or overhead lights. The soft light source illuminates the muzzle with a soft, diffused and glare-free light. In an embodiment, the inner surface of ULD 210 is disposed with a soft reflector (such as a reflective coating or a fabric) that reflects the light from the soft light source on to muzzle, thereby further increasing the incidence of soft light on the muzzle.
Thus LCA 208 mitigates the incidence of harsh light on the muzzle, thereby avoiding any glare. At the same time, it ensures that the muzzle is well-lit. This leads to good quality muzzle image captures in a wide variety of ambient lighting conditions.
Various alternatives or functional equivalents of LCA 208, including various types of light diffusers and soft light sources, are well-known and commonly used in photography and videography, and their details are not repeated herein in the interest of brevity. Further, diffusers and soft light sources using newer technologies may be developed over time. It will be apparent to a person skilled in the art that any such suitable alternatives may be used in conjunction with the present invention without deviating from its spirit and scope.
FIG. 2B shows an exploded view of the assembly of FIG. 2A. In an embodiment, trevis assembly 202 can be disassembled as shown in the figure. This makes it easier to transport trevis assembly 202, if needed. The figure shows trevis assembly 202 knocked down into a front section 250, a left section 252, a right section 254, an adjustable bar 256, four ground supports 258a-d, head rest 204, UDM 206, and LCA 208.
The illustrated collapsible design of trevis assembly 202 makes it possible to transport it on a relatively small vehicle, such as a three/four-wheeler, and deploy it quickly, inexpensively, and conveniently in any suitable location where ground supports 258a-d can be affixed to the ground 260 or to any strong and firm suitable base.
FIGs. 3A-3D illustrate trevis assembly 202 according to an embodiment of the present invention.
FIG. 3A shows a perspective view of trevis assembly 202 according to an embodiment of the present invention.
FIG. 3B shows a side view of trevis assembly 202 according to an embodiment of the present invention.
FIG. 3C shows a front view of trevis assembly 202 according to an embodiment of the present invention.
FIG. 3D shows a top view of trevis assembly 202 according to an embodiment of the present invention.
It will be apparent to a person skilled in the art that numerous modifications to the illustrated trevis assembly design can be made without deviating from the spirit and scope of the present invention.
NON-MUZZLE IMAGE REJECTION
In field conditions, while trying to capture a muzzle image, the user may accidentally or inadvertently capture images that do not contain the muzzle-print (hereinafter “non-muzzle images”). This may happen, for example, due to movement of animal 104 or of the camera of UD 106, or due to the user accidentally triggering image capture before satisfactory positioning has been achieved. Non-muzzle images may lead to wastage of computational and/or network resources if processed further by SMIS 102. Thus, in an embodiment, UD 106 is configured to automatically reject non-muzzle images, and not subject them to further processing. For this, UD 106 uses a preliminary classifier trained to distinguish muzzle images and non-muzzle images.
FIG. 4 shows example training images used to train a preliminary classifier according to an embodiment of the present invention. One or more sample muzzle images 402 are used as positive examples for the preliminary classifier. Similarly, one or more sample non-muzzle images 404 are used as negative examples for training the preliminary classifier. Thus trained, the preliminary classifier is used to identify any non-muzzle images captured in the future.
FIG. 5 shows a block diagram illustrating the training of a preliminary classifier for rejecting non-muzzle images according to an embodiment of the present invention. A set of training images 502, including muzzle images 402 and non-muzzle images 404, is processed by a feature extractor 504 to generate feature representations 506 of training images 502. In an embodiment, feature extractor 504 implements a Histogram of Oriented Gradients (HOG) algorithm and feature representations 506 comprise HOG matrices. In another embodiment, feature extractor 504 implements a Weber’s Local Descriptor (WLD) algorithm and feature representations 506 comprise WLD matrices.
Feature representations 506 are then labelled by a labeler 508 as positive and negative examples using human input. Human input regarding which training images 502 are muzzle images 402 and which are non-muzzle images 404 is recorded. Feature representations 506 corresponding to muzzle images 402 are labelled as positive examples, while those corresponding to non-muzzle images 404 are labelled as negative examples. Labeler 508 generates a labeled training data (LTD) 510 for a preliminary classifier 512.
LTD 510 is used to train preliminary classifier 512 in a supervised learning model. In an embodiment, preliminary classifier 512 is a Support Vector Machine (SVM) classifier. Thus trained, preliminary classifier 512 can then classify a new image as being a muzzle images or a non-muzzle image.
In various embodiments, SMIS 102 processes all newly acquired images during the registration process and/or the identification process as described above, for identifying and rejecting any non-muzzle images.
During the registration/training process, when SMIS 102 acquires a new image and detects it to be a non-muzzle image, the image is rejected from further processing and the user is suitably prompted to acquire another image. On the other hand, if the new image is detected to be a muzzle image, it is used further for the registration/training process to train the identification classifier.
During the identification/testing process, when SMIS 102 acquires a new image and detects it to be a non-muzzle image, the image is rejected from further processing and the user is suitably prompted to acquire another image. On the other hand, if the new image is detected to be a muzzle image, it is used further for the identification/testing process to identify the animal corresponding to the muzzle image.
ANIMAL REGISTRATION
FIG. 6 shows a block diagram illustrating registration of a new animal with SMIS 102. The user, via UD 106, requests registration of the new animal. On receipt of the request, SMIS 102 generates a Unique-ID (UID) for the new animal. A UID is a unique identifier assigned by SMIS 102 to each registered animal.
An image capture module (ICM) 602 in UD 106 (not shown) facilitates the user in capturing a muzzle image 604 of the new animal. Certain aspects of ICM 602 are described in more detail with reference to FIG. 9 and FIG. 10. In an embodiment, ICM 602 rejects any non-muzzle images as described with reference to FIG. 5. Upon confirming that muzzle image 604 is indeed a muzzle image, ICM 602 provides it to a feature extractor (FE) 606.
Feature Extraction
Feature extractor (FE) 606 computes a feature representation 608 of muzzle image 604. In an embodiment, FE 606 is embodied in UD 106. In another embodiment, FE 606 is embodied in IDS 108.
FE 606 applies a feature extraction algorithm to muzzle image 604 to build derived values (i.e. feature representation 608) intended to be informative and non-redundant, facilitating the subsequent machine learning steps. Any known feature extraction algorithm such as, but not limited to, Weber’s Local Descriptor (WLD), Scale Invariant Feature Transform (SIFT), Hough Transform, Gabor Feature Extraction, may be used in conjunction with the present invention. It will be apparent to a person skilled in the art that improved feature extraction algorithms developed in the future may also be used in conjunction with the present invention without deviating from its spirit and scope. In an embodiment, FE 606 applies the WLD algorithm and feature representation 608 includes a WLD matrix. In another embodiment, FE 606 applies the HOG algorithm and feature representation 608 includes a HOG matrix.
In a preferred embodiment, FE 606 applies a plurality of feature extraction algorithms to generate a plurality of feature representations 608, each of which is then processed as described herein.
Labeling
A UID labeler 610 labels feature representation 608 with the UID of the new animal currently being registered. The labeled feature representation 608 is added to a training data 612 and stored in AD 110. Similarly labeled feature representations of additional muzzle images of the new animal are also included in training data 612 and stored in AD 110. In an embodiment, between five (5) and fifteen (15) muzzle images of the new animal are thus processed and included in training data 612 at the time of registration.
In the foregoing manner, labelled feature representations of muzzle images of all registered animals are captured as training data 612 and maintained in AD 110 as a part of their respective animal identity profiles.
Classification
Training data 612 is used to train an identification classifier (IC) 614. IC 614 implements a classifier algorithm such as, but not limited to, a Support Vector Machine (SVM) algorithm, a K-nearest neighbor (K-NN) algorithm, a Fuzzy-KNN algorithm, a Decision Tree (DT), a Gaussian Mixture Model (GMM) algorithm, a Probabilistic Neural Network (PNN) algorithm, a Multilayer Perceptron (MLP) algorithm, a minimum distance (MD) algorithm, or a Naive Bayes classification model.
In an embodiment, identification classifier 614 is a Support Vector Machine (SVM) classifier. The SVM algorithm finds a hyperplane in an N-dimensional space (where N is the number of features) that distinctly classifies the data points. To separate the two classes of data points, there are many possible hyperplanes that could be chosen. The SVM classifier finds a hyperplane that has the maximum margin, i.e. the maximum distance between data points of both classes. Hyperplanes are decision boundaries that help classify the data points. Data points falling on either side of the hyperplane can be attributed to different classes. Also, the dimension of the hyperplane depends upon the number of features. If the number of input features is 2, then the hyperplane is a line. If the number of input features is 3, then the hyperplane is a two-dimensional plane. Support vectors are data points that are closer to the hyperplane and influence the position and orientation of the hyperplane. Using these support vectors, the SVM classifier maximizes the margin.
In various embodiments, identification classifier 614 provides a classification report which contains precision, recall and F1 score. Further, in various embodiments, SMIS 102 keeps track of one or more performance metrics of classifier 614 such as, but not limited to, accuracy, macro average, and weighted average. Accuracy is measure of how often identification classifier 614 makes the correct prediction. It is the ratio between the number of correct predictions and the total number of predictions. Macro average is calculation of metrics for each “class” independently and find the unweighted mean. A “class” refers to data points to be represented in vector space that have certain common characteristics. Weighted average is the calculation of metrics for each UID label, and find their average weighted by support. Weighted average is precision of all classes merged.
In various embodiments, multiple muzzle images of animal 104 are used to train identification classifier 614 at the time of registration in the manner described above. As the number of muzzle images used for training identification classifier 614 increases, initially the performance of identification classifier 614 improves significantly until it reaches a point of diminishing returns. The applicant has observed that the use of five (5) to fifteen (15) muzzle images is usually sufficient to train identification classifier 614 to perform adequately in most cases. It will be apparent to a person skilled in the art that more muzzle images (or less) may be used to train identification classifier 614 on a case-to-case basis without deviating from the teachings of the present invention.
In an embodiment, SMIS 102 captures and stores one or more secondary validation images of animal 104 at the time of registration, which are then used by SMIS 102 as described with reference to FIG. 11.
FIG. 7 shows example images captured at the time of registration of an animal 104 with SMIS 102 according to an embodiment of the present invention. As discussed earlier, SMIS 102 captures a set of muzzle images 702a-n of animal 104. In an embodiment, muzzle images 702a-n are used for training identification classifier 614 and are stored in AD 110 in the animal identity profile for animal 104, as described with reference to FIG. 6.
Additionally, in the illustrated embodiment, a secondary validation image (SVI) 704 is also captured. SVI 704 is an image of animal 104 that is suitable for visual identification of animal 104 by a human user of SMIS 102. SVI 704 is used for secondary validation of the biometric identification performed by SMIS 102. In other words, when SMIS 102 is testing a new muzzle image (as described with reference to FIG. 8) and identifies it to be that of animal 104, then SMIS 102 displays SVI 704 of animal 104 on UD 106. After viewing SVI 704, the user can confirm that it shows the same animal as the one the user is trying to identify, and thus validate the identification performed by SMIS 102.
As shown in the figure, SVI 704 is a front view of the entire body of animal 104. It will be apparent to a person skilled in the art that various other views of animal 104 may be used as SVI 704 including, but not limited to, side view of the full body of animal 104, front and/or side view of animal 104’s head, a view showing any distinguishing marks of the animal, and so on. In an embodiment, one or more secondary validation images of animal 104 are stored in its record in AD 110 and used for secondary validation in the manner described above.
ANIMAL IDENTIFICATION
FIG. 8 shows a block diagram illustrating identification of animal 104 with SMIS 102. ICM 602 in UD 106 (not shown) captures a muzzle image 804 of animal 104 to be identified. ICM 602 is described in more detail with reference to FIG. 9 and FIG. 10.
FE 606 computes a feature representation 808 of muzzle image 804. In an embodiment, FE 606 is embodied in UD 106. In another embodiment, FE 606 is embodied in IDS 108.
Identification classifier 614 attempts to identify animal 104 by comparing feature representation 808 with feature representations of registered animals. The feature representations of registered animals are stored in their respective animal identity profiles in AD 110 and included in training data 612 used to train IC 614. Thus trained, IC 614 evaluates feature representation 808 against training data 612 to identify the UID of the registered animal that closely resembles muzzle image 804.
Based on the foregoing comparison, identification classifier 614 computes an identification result 810.
In an embodiment, upon successful identification, identification result 810 includes the UID of the registered animal that most closely resembles muzzle image 804, and optionally, an indication of identification success. On the other hand, upon a failed identification, identification result 810 includes an indication of identification failure. In an embodiment, the failure is deduced if during secondary validation the user asserts that the identification is incorrect, as described with reference to FIG. 11.
In another embodiment, in case identification classifier 614 successfully matches feature representation 808 with those of a registered animal, result 810 is an identification success. If not, result 810 is an identification failure.
In an embodiment, in case of identification failure, SMIS 102 retries identification of animal 104 using a new muzzle image. Control returns to ICM 602 to capture another muzzle image, and the entire process is repeated. Thus, if the identification failure occurred due to poor quality of muzzle image 804, then the new muzzle image may result in identification success.
In an embodiment, SMIS 102 retries identification of animal 104 with new muzzle images for up to a pre-defined maximum number of retries. In another embodiment, SMIS 102 continues to retry identification of animal 104 until the user of UD 106 so desires. If all retries result in identification failure, SMIS 102 concludes that animal 104 is not registered with SMIS 102. In an embodiment, upon concluding that animal 104 is not registered, SMIS 102 proceeds to register animal 104 as described with reference to FIG. 6.
If identification result 810 is an identification success, then SMIS 102 concludes that animal 104 is a specific registered animal, hereinafter referred to as the identified animal. SMIS 102 conveys the UID of the identified animal to UD 106. It will be apparent to a person skilled in the art that any other modes of conveying the identity of the identified animal may be employed, without limitation, in conjunction with the present invention. In an embodiment, SMIS 102 also conveys one or more secondary validation images of the identified animal to UD 106. UD 106 displays the one or more secondary validation images of the identified animal to the user. This enables the user to validate that animal 104 has been correctly identified, by visually matching the appearance of animal 104 with the secondary validation image(s).
In various embodiments, SMIS 102 conveys the UID of the identified animal to one or more additional modules on UD 106, or IDS 108, or another device, for facilitating provision of various services in connection with the identified animal including, for example, accessing the animal’s medical record, owner information, location, diagnosis history, vaccination history, pregnancy status, list of vaccinations due, and so on.
SHORTLISTING PARAMETERS
As the number of animals registered with SMIS 102 grows, the time taken for identification by identification classifier 614 increases, thereby slowing down the identification process. Further, as the number of registered animals becomes large, the likelihood of finding two different animals with similar muzzle-prints increases. This can result in incorrect identification. To mitigate at least the foregoing issues, in an embodiment, SMIS 102 includes a shortlisting module to shortlist the registered animals to be considered during the identification process based on one or more shortlisting parameters. In various embodiments, the shortlisting parameters include, but are not limited to, the owner of the animal, the location of the animal, breed and age.
Owner
In an embodiment, UD 106 includes a user identity module configured to provide an identity of the user, which is used in animal identification as described herein.
In an embodiment, SMIS 102 records ownership information of each animal at the time of registration. In an embodiment, the ownership information is entered by the user at the time of registration. In another embodiment, SMIS 102 infers the ownership information from the identity of the user registering the animal. The identity of the user can be defined by the login of the user, contact information of the user (such as email address and/or phone number), or by other means known in the art. Each user is associated with one or more animal owners either explicitly based on data entered in SMIS 102, or implicitly based on past activity of the user.
When a user requests identification of an animal, SMIS 102 initially assumes that the animal must belong to one of the animal owners that the user is associated with. SMIS 102 short-lists only those registered animals that are owned by the one or more owners associated with the user.
Location
In an embodiment, UD 106 includes a location module configured to detect a geographical location of the user device, which is used in animal identification as described herein.
In an embodiment, UD 106 includes a location module capable of identifying the geographical location of UD 106, such as a Global Positioning System (GPS) module. ICM 602 is configured to record the location of each muzzle image capture, for example, within the metadata associated with the muzzle image, such as geotags within the Exchangeable Image File format (EXIF) data of the muzzle image. SMIS 102 uses the location at which the muzzle images of an animal were captured at the time of registration of that animal to identify the location of registration of the animal. This location is stored in AD 110 as a part of the record of the registered animal.
When a new muzzle image is received for identification/testing, SMIS 102 uses the location at which the new muzzle image was captured to shortlist only those registered animals that were registered within an acceptable distance of the location of the new muzzle image, for example within one kilometer of the location. Identification classifier 614 only considers these short-listed registered animals as identification candidates.
In an embodiment, SMIS 102 sorts the short-listed animal records by the distance between location of registration and location of the new muzzle image. Identification classifier 614 evaluates the new muzzle image against registered animals in this order, i.e. animals registered closest to the location of the new muzzle image are evaluated first.
It will be apparent to a person skilled in the art that various combinations of multiple short-listing parameters can be used in conjunction with the present invention without deviating from its spirit and scope.
IMAGE CAPTURE
Image capture module 602 ensures that good quality muzzle images, which are suitable for use in biometric identification, are obtained from UD 106. To this end, ICM 602 assists the user of UD 106 in capturing muzzle image and, in various embodiments, also performs pre-processing of the muzzle images to facilitate better performance of FE 606 and identification classifier 614.
FIG. 9 shows an example user interface for guiding the user during capture of muzzle image 604, according to an embodiment of the present invention. The figure shows a display 902 of UD 106. In an embodiment, display 902 is a touch-sensitive display, or a touchscreen, as shown in the figure. Display 902 shows a user interface 904 of image capture module 602. UD 106 presents interface 904 whenever the user wishes to identify an animal. Interface 904 includes a camera view area 906 which displays the view before a camera of UD 106 in real time. In an embodiment, ICM 602 sets the resolution of the camera to a pre-defined resolution range, such as within the range 300-340 pixels on the x-axis by 350-380 pixels on the y-axis. Using this resolution range, the uniqueness of muzzle (beads and ridges) nostrils and this numeric range frame captures desired area from predefined distance.
The user adjusts the position, yaw, pitch, and roll of UD 106 while viewing camera view area 906 to adjust the view in a manner that the central portion of camera view area 906 is occupied by the muzzle, as shown. Further, interface 904 includes a guidance box 908 for guiding the user during image capture in a manner that the obtained muzzle image 604 is of good quality and suitable for use in biometric identification. The user adjusts the position, yaw, pitch, and roll of UD 106 in a manner that the muzzle print area (i.e. the muzzle area between the nostrils, as shown) substantially fills box 908 and at the same time is substantially enclosed by it.
Once the user has achieved satisfactory positioning as described above, he/she captures an image using an image capture element 910. In various embodiments, element 910 can be a soft button on the touchscreen of UD 106 (as shown in the figure), a physical input element on UD 106 such as a button, a gesture recognized by UD 106 such as a tap, double-tap, long-press, etc., or any other suitable user-interface element known in the art or developed in the future.
The area within box 908 is captured as muzzle image 604. In an embodiment, ICM 602 scales muzzle image 604 to pre-defined standard dimensions, such as 100x300 pixels. Resizing into 100x300 pixels results in small WLD matrices without affecting the uniqueness, saving the storage, faster processing time.
FIG. 10 shows example images at various steps during the muzzle image capture process according to an embodiment of the present invention. A first image 1002 shows the view as seen in camera view area 906 at the time of image capture. ICM 602 auto-crops first image 1002 to retain only the portion within box 908, as seen in a second image 1004.
Further, ICM 602 resizes second image 1004 to a pre-defined dimension of 100x300 pixels an obtains a third image 1006 to be used as muzzle image 604 for further processing by FE 606. It will be apparent to a person skilled in the art that various other pre-defined dimensions may be advantageous with different feature extractors. For example, with a HOG feature extractor, it is preferable to resize the second image 1004 to 256x256 pixels for preparing third image 1006.
SECONDARY VALIDATION
FIG. 11 shows screen representations illustrating an interface for secondary validation according to an embodiment of the present invention. Once SMIS 102 identifies an animal, it displays a secondary validation image of the animal to the user and allows the user to validate if the animal has been identified correctly. A secondary validation interface 1102 on display 902 of UD 106 includes a secondary validation image 1104 of the identified animal. In an embodiment, interface 1102 includes an animal details section 1106 that displays additional details of the identified animal, such as its name. The user can visually compare the identified animal with image 1104 and/or use the details in section 1106 to confirm that the identification has been done correctly.
If SMIS 102 has correctly identified the animal, the user confirms the identification using a confirmation user-input element 1108. On the other hand, if the identification is incorrect, the user rejects the identification using a rejection user-input element 1110.
In an embodiment, after the user rejects the identification, SMIS 102 retries the identification using the same muzzle image but with a different feature extraction algorithm. For example, if identification using a WLD feature extractor leads to incorrect results, SMIS 102 retries identification of the same muzzle image using a HOG feature extractor.
In an embodiment, after the user rejects the identification, SMIS 102 prompts the user to acquire a new muzzle image and restarts the identification process with the new muzzle image.
In various embodiments, elements 1108 and/or 1110 can be a soft button on the touchscreen of UD 106 (as shown in the figure), a physical input element on UD 106 such as a button, a gesture recognized by UD 106 such as a tap, double-tap, long-press, etc., or any other suitable user-interface element known in the art or developed in the future.
FIG. 12 shows the distribution of responsibilities between UD 106 and IDS 108 according to an embodiment of the present invention. ICM 602, FE 606, and identification classifier 614 reside on UD 106, which communicates IR 810 to IDS 108 over network 112.
FIG. 13 shows the distribution of responsibilities between UD 106 and IDS 108 according to another embodiment of the present invention. ICM 602 resides on UD 106 and communicates MI 804 to IDS 108 over network 112. IDS 108 includes FE 606 and classifier 610.
USER DEVICE
FIG. 14 shows an exemplary user device 106 in accordance with an embodiment of the invention. User device 106 comprises a camera 1402 which includes an image sensor, a global positioning system (GPS) sensor 1404 to determine a geographical position of user device 106, a display unit 1406 configured to display a user interface, and a user input unit 1408 configured to receive a user input from a user of the user device. In various embodiments, user input unit 1408 can include a touch screen, a keyboard, a gesture recognition device, and so on.
User device 106 further includes a memory 1410 electrically connected to the image sensor and configured to store an image captured by the image sensor, a processor 1412 electrically connected to the image sensor, GPS sensor 1404, display unit 1406, user input unit 1408, and memory 1410. Processor 1412 is configured to control the user interface on display unit 1406, receive the user input from user input unit 1408, receive the geographical position of user device 106 from GPS sensor 1404, and to operate at least one of display unit 1406, the image sensor, GPS sensor 1404, and memory 1410 based on the user input.
User device 106 further includes a network interface unit 1414 electrically connected to processor 1412 and memory 1410. Network interface unit 1414 is configured to provide data network connectivity to the user device.
In various embodiments, UD 106 can be a smart phone, a tablet, a personal digital assistant (PDA), a laptop or desktop with a camera, a dedicated device, or a combination of the foregoing. It will be apparent to a person skilled in the art that a variety of devices and / or device combinations may be used to achieve the functionality of UD 106 as set forth in this description, and that any of those may be employed in conjunction with the present invention without limitation.
Further, the present invention is not to be limited in scope by the specific embodiments described herein. It is fully contemplated that other various embodiments of and modifications to the present invention, in addition to those described herein, will become apparent to those of ordinary skill in the art from the foregoing description and accompanying drawings. Thus, such other embodiments and modifications are intended to fall within the scope of the following appended claims.
Further, although the present invention has been described herein in the context of particular embodiments and implementations and applications and examples and in particular environments, those of ordinary skill in the art will appreciate that its usefulness is not limited thereto and that the present invention can be beneficially applied in any number of ways and environments for any number of purposes. Accordingly, the claims set forth below should be construed in view of the full breadth and spirit of the present invention as disclosed herein.