Claims:WE CLAIM:
1. A system (105) for low power artificial intelligence on audio video clip from audio video source (110), the system characterized in that comprising:
a source health classifier module (140) and a source frame timer module (155), for selecting audio video frame, wherein selection is based on relevance of frame;
a data frame filter module (160) and a statistical compute module (165), for analysing the frame, wherein noise free frame with adequate motion is selected;
an artificial intelligence module (175), for generating inferences, wherein the artificial intelligence module (175) has optimized neural net implementation to operate on low power devices.
2. The system (105) as claimed in claim 1, wherein health classifier module (140) classifies the video clip received from the video source (110) as valid or invalid, based at least on a decision of a classification decision model (145) and only video clips classified as valid are further processed.

3. The system (105) as claimed in claim 1, wherein the source frame timer module (155) selects right audio video frame by providing enough time gap between last frame and current frame, based at least on a decision of a frame selection decision model (150) and only qualified audio video frames are further processed.

4. The system (105) as claimed in claim 1, wherein the data frame filter module (160) and the statistical compute module (165) filters audio video frame and further finding magnitude of relative motion based on at least on a decision of a compute support model (170) and only audio video frames with relevant relative motion are further processed.

5. The system (105) as claimed in claim 1, wherein the artificial intelligence module (175) generates inferences from audio video clips, with neural nets optimized for low power devices with lesser number of parameters, using depth wise separable convolution.

6. A method (200, 250) for low power artificial intelligence on audio video clip from audio video source (110), using system (105) as claimed in claim 1, the method (200, 250) characterized in that comprising::
selecting audio video frame, by the source health classifier module (140) and the source frame timer module (155), wherein selection is based on relevance of frame;
analysing the frame, by the data frame filter module (160) and the statistical compute module (165), wherein noise free frame with adequate motion is selected;
generating inferences, by the artificial intelligence module (175), wherein the artificial intelligence module (175) has optimized neural net implementation to operate on low power devices.
7. The method (200, 250) as claimed in claim 6, further comprising classifying the video clip received from the video source (110) as valid or invalid, by a health classifier module (140), wherein the video clip is classified based at least on a decision of a classification decision model (145) and only video clips classified as valid are further processed.

8. The method (200, 250) as claimed in claim 6, further comprising of selecting right audio video frame by providing enough time gap between last frame and current frame, by a source frame timer module (155), where in data frame is selected at least on a decision of a frame selection decision model (150) and only qualified audio video frames are further processed.

9. The method (200, 250) as claimed in claim 6, further comprising of filtering audio video frame and further finding magnitude of relative motion, by the data frame filter module (160) and the statistical compute module (165), where in magnitude of motion is arrived at least on a decision of a compute support model (170) and only audio video frames with relevant relative motion are further processed.

10. The method (200, 250) as claimed in claim 6, further comprising of generating inferences from audio video clips, by the artificial intelligence module (175), with neural nets optimized for low power devices with lesser number of parameters, using depth wise separable convolution.
, Description:FIELD OF THE INVENTION
[0001] The present disclosure relates to low power artificial intelligence. In particular, the present disclosure relates to using audio video artificial intelligence run on a system with low compute resources.
BACKGROUND OF THE INVENTION
[0002] Video Surveillance system widely uses Artificial Intelligence to analyse the audio and images in order to recognize humans, vehicles, objects and events. The Artificial Intelligence requires dedicated high amount of compute power to work either as CPU (Central Processing Unit) or GPU (Graphics Processing Unit). Therefore, it is important to have an approach to have Artificial Intelligence run on low power or less compute power systems.

[0003] Video surveillance systems with Artificial Intelligence, however, require high amount of compute power. As a result, overall cost of system will be high to cover BOM (Bill of Materials) related to higher computer power either as CPU or GPU. In many locations, such as locations in developing countries affordability index of large-scale video surveillance system with Artificial Intelligence reduces because of higher BOM cost. Therefore, there is a need for a system and a method for using Artificial Intelligence in low compute systems with lesser BOM cost.
BRIEF SUMMARY OF THE INVENTION
[004] This summary is provided to introduce a selection of concepts in a simple manner that are further described in the detailed description of the disclosure. This summary is not intended to identify key or essential inventive concepts of the subject matter nor is it intended for determining the scope of the disclosure.

[005] An example of a method for low power artificial intelligence is disclosed. This method comprises of selecting audio video data from a source which is not having any issues, by source health classifier module. Further, the method comprises of selecting right frame, by source frame timer module and further filtration by data frame filter module. Further, the method comprises of statistical computations to finalize and quality frame by statistical compute module and further to artificial intelligence module.

[006] An example of a system for low power artificial intelligence is disclosed. This system comprises of selecting audio video data from a source which is not having any issues, by source health classifier module. Further, the system comprises of selecting right frame, by source frame timer module and further filtration by data frame filter module. Further, the system comprises of statistical computations to finalize and quality frame by statistical compute module and further to artificial intelligence module.

[007] To clarify advantages and features of the present disclosure further, a more particular description of the disclosure is rendered by reference to specific embodiments thereof, which is illustrated in the appended figures. It is to be appreciated that these figures depict only typical embodiments of the disclosure and are therefore not to be considered limiting of its scope. The disclosure is described and explained with additional specificity and detail with the accompanying figures.

BRIEF DESCRIPTION OF THE FIGURES
[008] The disclosure is described and explained with additional specificity and detail with the accompanying figures in which:

[009] FIG. 1A illustrates an environment of a system for low power artificial intelligence, in accordance with one embodiment of the present disclosure;

[0010] FIG. 1B illustrates functional block diagrams of the system, in accordance with one embodiment of the present disclosure; and

[0011] FIGS. 2A and 2B shows a method for low power artificial intelligence, in accordance with one embodiment of the present disclosure.

[0012] Further, persons skilled in the art to which this disclosure belongs may appreciate that elements in the figures are illustrated for simplicity and may not have been necessarily been drawn to scale. Furthermore, in terms of the construction of the system and, one or more components of the system may have been represented in the figures by conventional symbols, and the figures may show only those specific details that are pertinent to understanding the embodiments of the present disclosure so as not to obscure the figures with details that are readily apparent to those of ordinary skill in the art having benefit of the description herein.

DETAILED DESCRIPTION
[00013] For the purpose of promoting an understanding of the principles of the disclosure, reference is now made to the embodiment illustrated in the figures and specific language is used to describe them. It should nevertheless be understood that no limitation of the scope of the disclosure is thereby intended. Such alterations and further modifications to the disclosure, and such further applications of the principles of the disclosure as described herein being contemplated as would normally occur to one skilled in the art to which the disclosure relates are deemed to be a part of this disclosure.

[00014] It may be understood by those skilled in the art that the foregoing general description and the following detailed description are exemplary and explanatory of the disclosure and are not intended to be restrictive thereof.

[00015] The terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process or method that comprises a list of steps does not include only those steps but may include other steps not expressly listed or inherent to such a process or a method. Similarly, one or more devices or sub-systems or elements or structures or components preceded by "comprises... a" does not, without more constraints, preclude the existence of other devices, other sub-systems, other elements, other structures, other components, additional devices, additional sub-systems, additional elements, additional structures, or additional components. Appearances of the phrase “in an embodiment”, “in another embodiment” and similar language throughout this specification may, but do not necessarily, all refer to the same embodiment.

[00016] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs. The system, methods, and examples provided herein are illustrative only and not intended to be limiting.

[00017] The present disclosure relates to a system and a method for low power artificial intelligence on one or more audio video sources. More specifically, the system and method relate to classifying validity of received frame from audio video source and further picking right frame based on source dependent timer. Upon receiving right frame, the system applies data frame filters and subsequently does statistical computations prior to processing in artificial intelligence. Thus, this system and method ensures not all but right data frame is processed by artificial intelligence thereby reducing processing needs.

[00018] Embodiments of the present disclosure are described below in detail with reference to the accompanying figures.

[00019] Referring to FIG. 1A, a system 105 for low power artificial intelligence from one or more audio video sources 110 is shown, in accordance with one embodiment of the present disclosure. The one or more audio video sources 110 includes, but is not limited to, a Digital Video Recorder (DVR), a Network Video Recorder (NVR), an IP (Internet Protocol) camera or any other audio video streaming device known in the art. In one example, when the video source 110 is a DVR, an NVR or an IP camera, the audio video source 110 records a video clip of the surroundings, in a digital format with the help of a motion analysis software, upon detection of motion in the video. In another example, the video source 110 may comprise passive infrared sensors to detect motion based on body heat. Based on the output of the passive infrared sensors, the audio video source 110 may start recording the video clip.

[00020] When the audio video source 110 records a video clip comprising an event of interest, the video clip is sent to the system 105 for further processing.

[00021] The system 105 may comprise at least one processor 120, a memory 125 and an I/O interface 130. The at least one processor 120 may be implemented as one or more microprocessors, microcomputers, microcontrollers, digital signal processors, central processing units, state machines, logic circuitry, and/or any devices that manipulate signals based on operational instructions. Among other capabilities, the at least one processor 120 is configured to fetch and execute computer-readable instructions stored in the memory.

[00022] The memory 125 may include any computer-readable medium known in the art including, for example, volatile memory, such as static random access memory (SRAM) and dynamic random access memory (DRAM), and/or non-volatile memory, such as read only memory (ROM), erasable programmable ROM, flash memories, hard disks, optical disks, and magnetic tapes.

[00023] The I/O interface 130 may include a variety of software and hardware interfaces, for example, a web interface, a graphical user interface, and the like. Further, the I/O interface 130 may enable the system 105 to communicate with other computing devices, such as web servers and external data servers (not shown). The I/O interface 130 may facilitate multiple communications within a wide variety of networks and protocol types, including wired networks, for example, LAN, cable, etc., and wireless networks, such as WLAN, cellular, or satellite. The I/O interface 130 may include one or more ports for connecting a number of devices to one another or to a server. The I/O interface 130 receives the audio video data from the audio video source 110 and the processor 120 processes the audio video data using computer program instruction stored in the memory 125.

[00024] Although the present disclosure is explained by considering that the system 105 is implemented on a server, it may be understood that the system 105 may also be implemented in a variety of computing systems, such as a mainframe computer, mobile, single board computer and the like.

[00025] Referring to FIG. 1B, functional blocks of the processor 120 is shown, in accordance with one embodiment of the present disclosure. The processor 120 comprises a source health classifier module 140, supported by classification decision model module 145, a source frame timer module 155, supported by frame selection decision model module 150, a data frame filter module 160, a statistical compute module 165, supported by compute support model 170, and an artificial intelligence module 175. The functioning of each of the modules is explained in detail below with reference to FIGS. 2A and 2B, in conjunction with FIGS. 1A and 1B.

[0026] Referring to FIGS. 2A and 2B, in conjunction with FIGS. 1A and 1B, a method 200 and 250 for selecting a right data frame based on time slot and further filtering, statistical compute prior to artificial intelligence, in accordance with one embodiment of the present disclosure. The method 200 and 250 may be described in the general context of computer executable instructions. Generally, computer executable instructions may include routines, programs, objects, components, data structures, procedures, modules, functions, etc., that perform particular functions or implement particular abstract data types. The method 200 and 250 may also be practiced in a distributed computing environment where functions are performed by remote processing devices that are linked through the network. In a distributed computing environment, computer executable instructions may be located in both local and remote computer storage media, including memory storage devices.

[0027] The order in which the method 200 and 250 is described and is not intended to be construed as a limitation, and any number of the described method blocks can be combined in any order to implement the method 200 and 250 or alternate methods. Additionally, individual blocks may be deleted from the method 200 and 250 without departing from the scope of the disclosure described herein. Furthermore, the method may be implemented in any suitable hardware, software, firmware, or combination thereof. However, for ease of explanation, in the embodiments described below, the method 200 and 250 may be implemented in the above-described system 105.

[00028] Each of the video sources 110 is assigned a device priority, based on a location of the video source 110, health of the video source 110, and so on. In one example, the device priorities are assigned manually by an operator. In another example, the device priorities are determined by the video source 110 based factors including, but not limited to, a location of the video source 110, a health of the video source 110, and so on. When the video source 110 records a video clip comprising an event of interest, the video clip is sent to the system 105 for further processing.

[00029] At step 205, the source health classifier module 140 receives a clip from the audio video source 110.

[00030] At step 210, the source health classifier module 140 determines whether the audio video source 110 is one of a healthy source and a suspect source, using a classification decision model 145. The classification decision model 145 may be dynamically modelled at remote server or the system 105 based on a plurality of first parameters. In one example, the first parameter may be related to a health information of the video source 110. The health information may be associated with a hardware of the video source 110 or a quality of video clip recorded by the video source 110. In one example, the health information is determined based on various factors that may contribute to noise in the video clip, such as electrical disturbances, loose connections, cobwebs, poor infrared quality of the video source 110, and so on. For example, low intensity flickering of IR may erroneously trigger recording of the video clip by the Digital Video Recorder (DVR) / Network Video Recorder (NVR).

[00031] In one implementation, the health information of the video source 110 is determined using device analytics. For example, the system may perform device analytics by sampling video clips captured by the video source 110 and analysing the video clips for identifying periodic disturbances such as flickering. In another implementation, the system may perform the device analytics of the video source 110, with the help of another device which includes a software agent that monitors functioning of the video source 110. Further to the above, if the video source 110 is determined to have a performance or functional problem or inadequacy, an operator or monitoring person raises a ticket for a vendor of the video source 110 to solve the problem. If the video source 110 is awaiting resolution of the problem, a status of the ticket is also considered during classification of the video source 110 as a suspect source or a healthy source, by the classification decision model 145.

[00032] The source health classifier module 140, further transmit only healthy audio video source 110 data frames to the source frame timer module 155.

[00033] At step 215, the source frame timer module 155, determines if there is enough time gap between last data frame and current frame, using a frame selection decision model 150. The frame selection decision model 150 may be dynamically modelled or by the system 105 based on plurality of second parameters. This step is referred as frame index verification process.

[00034] In one implementation, the second parameter related to a frame selection depends on type of artificial intelligence processing needed. For example, a crowd monitoring frequency set determines the gap between data frames. In another implementation, the second parameter related to frame section depends on location of audio video sources 110. For example, frame selection model can skip frames from inside audio video sources, if outside audio video sources haven’t detected anything. In another implementation, the second parameter related to frame selection depends on available computing power dynamically built as a function of system processor and memory usage.

[00035] As at step 216, the source frame timer module 155, further transmit data frames that meets criteria to the data frame filter 160.

[00036] At step 220, the data frame filter module 160, applies filters on selected frame. In one implementation, noise filters are applied on selected frame. In another implementation, gaussian filters are applied.

[00037] The data frame filter module 160, will ignore frame that are below defined threshold after filtration.

[00038] At step 225, the statistical compute module 165, finds out magnitude of relative motion or change in data frame, using the compute support model 170. In one implementation, the statistical compute module 165, finds out the magnitude of motion in current data frame with respect to last processed frame information from the compute support model 170. In one implementation, the threshold for comparison is arrived based on available computing power dynamically built as a function of system processor and memory usage. In another implementation, the threshold is taken as user input.

[00039] The statistical compute module 165, will ignore the frame has less relative motion, otherwise data frame will reach artificial intelligence module 175.

[00040] As at step 226, the artificial intelligence module 175 receive data after statistical operations for generating inferences. The artificial intelligence module 175 is optimized neural net implementation for low power devices. In one implementation, the artificial intelligence module 175, uses depth wise separable convolution resulting in lesser number of parameters. In another implementation neural net is training with limited number of classes focussed for the purpose.

[00041] The present invention classifies validity of received frame from audio video source and further picks right frame based on source dependent timer. Then, upon receiving right frame, the system applies data frame filters and subsequently does statistical computations prior to processing in optimized artificial intelligence module. Thus, system and method ensures not all but right data frame is processed by artificial intelligence thereby reducing need of heavy processing and hence lower device can be used for artificial intelligence inferences.

[0042] While specific language has been used to describe the disclosure, any limitations arising on account of the same are not intended. As would be apparent to a person skilled in the art, various working modifications may be made to the method in order to implement the inventive concept as taught herein.

[0043] The figures and the foregoing description give examples of embodiments. Those skilled in the art will appreciate that one or more of the described elements may well be combined into a single functional element. Alternatively, certain elements may be split into multiple functional elements. Elements from one embodiment may be added to another embodiment. For example, orders of processes described herein may be changed and are not limited to the manner described herein. Moreover, the actions of any flow diagram need not be implemented in the order shown; nor do all of the acts necessarily need to be performed. Also, those acts that are not dependent on other acts may be performed in parallel with the other acts. The scope of embodiments is by no means limited by these specific examples. Numerous variations, whether explicitly given in the specification or not, such as differences in structure, dimension, and use of material, are possible.