Description:Field of Invention
Now a day’s traffic control is the major issue in most of the cities. To minimize this issue different efforts have been applied. One such effort is use of pre determined timer circuits to control traffic signals. Despite the many benefits of traffic signals it has most of the restrictions for the current traffic. Most of the vehicles are made to wait at junctions even though there is less or no traffic. To overcome the problems of traditional traffic systems an Intelligent Transport system is developed. Video detection plays an important role in transportation system. In this approach also detection and tracking is difficult because of change in the environment, shadows, lighting etc. To improve the performance of vehicle detection under different conditions, In Intelligent Transport System vehicle identification depends on the movement division with dynamic foundation and unique changing calculation to accomplish feature space division.
Background of the Invention
Due to heavy increase of traffic the current traffic signal mechanism does not plays a key role. To overcome this problem an Intelligent Transport System is designed. This ITS uses different computing techniques, vision advancements, different communication approaches to deal with different types of transportation issues. The basic versions of ITS leads many problems in terms of setup cost, limited inclusion. To overcome these problems camcorders is designed and setup at public places. To improve the performance of traffic surveillance detection of moving vehicles is most important thing.
Before designing the algorithm, we set some thresholds for effective design of moving vehicles. Those are, vehicles are accurately identified even from the background, detected on different traffic conditions, at different speed and different environmental situations like cloudy, shadow, rain etc.
For designing a technique for detection of moving vehicle with the above conditions faces lot of challenges. Some of the challenges are, noise in the video affect the quality of detection, changing the position of vehicles in the traffic conditions, changes in air quality, changes in the background conditions, detection of vehicles without lights, detection of vehicles in shadow, calculating speed and time of travel(WO2017078072A1).
Many number of approaches are designed to satisfy the above conditions and among them the basic methods are background subtraction and optical flows (CN106373394B). In this back ground subtraction is mainly used in environmental changes situations like ambient lighting, and changes in the background of vehicles. Optical flows technique is mainly used for detecting vehicle in noisy and disruption condition. Again these two approaches lead to many problems and to overcome these problems and detecting moving vehicle accurately a novel approach optimal probabilistic Neural Network is developed.

Summary of the Invention
To detect the moving vehicles with high resolution rate a technique OCS-PNN is designed. In this technique the CS is hybridized with OLB for optimizing the weight value of the PNN model. This model is implemented in two modules. One is Generation of background and other one is detection of moving vehicle using OCS-PNN model. In this technique to improve the performance of detection for PNN classifier the weights are optimally selected by using OCS algorithm. In this OCS-PNN approach we are facing some of the difficulties like detection of vehicle by using passive optical sensors. The problem occurs if the vehicles are differing in their size, shape and shading. To resolve these issues a novel approach using artificial neural networks is introduced. It detects the moving vehicle in two stages. First stage is Hypothesis generation and in second stage verification of generated hypothesis using Histogram orientation Gradients. Simply the histogram orientation gradients are given to given to OANN classifier and weights are selected using Grasshopper optimization algorithm.
Brief Description of Drawings
Figure 1: Block Diagram of OCS-PNN Model
Figure 2: Execution Moving Vehicle detection architecture using Artificial Neural Networks.
Figure 3: Hypothesis Verification Process
Detailed Description of the Invention
To detect the moving vehicles effectively a novel approach OCS-PNN is developed shown in fig 1. This model consists of two blocks. 1. Back ground generation 2. Moving vehicle detection. In back ground generation make an assumption of the color images are processed which contains multiple back grounds in real tie environment. Quality of the images are improved by using color images and have to adapt brightness , low coloring conditions and multiple background colors. The main problem in background generation module is external memory access is inefficient. To overcome this problem the first step in the algorithm is detailed analysis of memory for different models . For example Single Gaussian model uses single background model and produces the outputs the represents the difference between clustering and Multiple Gaussian techniques. For all the techniques the following quantities are estimated. Those are pixel dimension of the model, location dimension of the model, size of the cluster, complete model size. The background models contain two attributes. The first attribute is dimensions of the complete model for test method and second attribute is location model size. During realization of hardware the model is read and written with the speed of frequency of clock. If the resolution of the video increase it requires more memory. To satisfy all these requirements clustering approach is used for complex background images. Next step is extraction of moving vehicles from simple background model which are enhanced by background update procedure. After completion of background generation, by using OCS-PNN algorithm moving vehicles are detected from the videos. This moving vehicle detection stage consists of three processes as Block estimation, vehicle detection process and background updating process. Block calculator algorithm is used to estimate the block statistics for minimizing the unwanted blocks in the background area.
The main objective of this approach is, first to get better precision OCS algorithm is used for finding optimum weights for PNN. Second an OBL scheme is used for finding opposite number of solutions and by using this distance of data from the optimal solution is decreased. Third to identify the moving vehicle completely background model is mapped to incoming pixels. PNN is slight variation of radial basic function and also approximation of Bayesian technique. PNN depends on Parzen probabilistic density for arrangement of source vector with target class of the issue.PNN is represented as a 3-layered feed forward system with input layer, hidden layer, and output layer. In this input layer is used to transfer the source samples to radial basis layer. Hidden layer is used to calculate the distance between the input and the training vector for generating samples. Output layer is used to add the contribution of all the inputs and generate yield vector in probability function. Finally, a competitive transfer function is generated at the middle of first and the third layers to derive a class with high probability of it become correct. OBL is used as an effective method for differential evolution to optimization problems. In the process of solution evaluation for a problem, in some scenarios opposite solution may yields best results. By using this opposite solution the distance for an object is decreased from the optima. This OCS-PNN model considers the first stage as input layer. Assume it consists of n number of neurons. In this training test set each neuron acts as separate entity. Now from the given input neurons a pattern neuron is calculated by simply multiplying the input neuron with weight. The weights of each neuron are calculated using OCS algorithm. The final layer is the output layer and this layer only one output class is available.
To overcome some of the issues in OCS-PNN approach a novel Artificial Neural Network is introduced. In this the moving vehicle is identified in two steps shown in Fig 2. In the first step Hypothesis is generated for the vehicles under shadow and in second step verification of hypothesis generated from the first step and features are to be extracted using HOG algorithm. The main purpose of Hypothesis generation is extracting the hypothesis of vehicles from the images taken from roads. In this step vehicles are extracted from the shadow region appears under the vehicles. Hypothesis is generated based on three facts such as texture, shadow and symmetry. The generation process starts by identifying the shadows under the vehicles. After identification of shadows the regions above the shadows are treated as ROI. In this identification the main important factor is gray level of the shadow varies for different days. So, it is necessary to fix a threshold value for identification of shadow. For fixing the threshold no need of minimum limit for intensity but an upper boundary has to be fixed for the intensity value. The value of the threshold mainly depends on color and illumination level of the surface of the road and it does not depends on the location of the road. After detection of the vehicle apply the Sobel operation for extracting the edges of the ROI and then areas above the edges are removed. To control length of line segments, the lines that belong to the vehicles are separated and the remaining portions are considered as background and the region above the shadow is considered as ROI. After completion of extraction apply morphological method to makes the shadow more noticeable. After completion of Hypothesis generation the extricated hypothesis are send to hypothesis verification stage. This steps verifies the hypothesis and make a conclusion that whether they are vehicles or not by using classification algorithms. The hypothesis verification process is shown in fig 3. This stage consists of three steps. 1. Feature extraction 2. Training using OANN model 3. Verification process. For feature extraction HOG is used. HOG descriptor is most popular features in the vision dependent target object detection. This method calculates the occurrence of gradient in the local part of the image. By using HOG the image is segmented into small parts called cells. HOG is calculated for each pixels in the cell. After features are extracted using HOG the features are send to OANN classifier to train the combined feature vectors. After completion of training using OANN model the features are sent to verification process. In this verification process the objects are checked whether they are vehicles or no vehicles.
After extraction of features those are trained by using OANN classifier and after that weights are selected using by using GOA. After completion of all the training the objects are send to OANN classifier which will give the results that it is vehicle or not. The main intention of this technique is identification of surrounding vehicles in the pictures from the camera. OANN is mainly used for division of inputs into two categories like vehicles or not vehicles. In this approach also neural network is categorized as three layers as input, hidden and output layers and it can increase the number of layers depending upon the problem we are trying to solve. ANN consists of collection of neurons which are interconnected among themselves and al the links consist of weights specify the strength. In this approach to reduce the error value weights are calculated using GOA. The verification stage checks whether the object obtained from the first step is detected as vehicle or not.
4 Claims & 3 Figures , Claims:The scope of the invention is defined by the following claims:

Claim:
1. The System/Method to Detect Moving Vehicle using Artificial Neural Networks comprising the steps of:
a) An optimal probabilistic neural network model to detect the moving vehicle.
b) An auto monitoring system to identify moving object recognition, motion segmentation, object tracing, object classification etc.
c) The automated system will overcome the problems of traditional approaches like objects under shadows, environmental changes, different shapes of objects etc.
2. The System/Method to Detect Moving Vehicle using Artificial Neural Networks as claimed in claim1, to identify the object detection, classification and tracing of objects an optimized OCS-PNN model is designed.
3. The System/Method to Detect Moving Vehicle using Artificial Neural Networks as claimed in claim1, to improve the speed of detection the weight matrix of PNN is combined with Oppositional Cuckoo search algorithm is designed.
4. The System/Method to Detect Moving Vehicle using Artificial Neural Networks as claimed in claim1, again to improve the accuracy of detection optimal Artificial neural Network based algorithm is designed.