Modified Neural Network-based Object Classification in Video Surveillance System

Abstract

Visual surveillance emerged as an active automated research area of Computer Vision from the traditional mathematical approach to neural networks. A novel modified neural network technique for object detection and classification for input images and video feed from many cameras overlapping target areas is presented in this research.Modified Neural Network methodology represents layered architecture as the input, preprocessing and Operation layer, to simplify the processing needed to prepare for training neural networks. This strategy aids in delegating the tasks to layers with predefined tasks thus simplifying training, reducing computational requirements, and delivering performance. Two modules of the Neural Network will process the input. The first module is a modified Neural Network and will differ from traditional Neural Network in respect of connectivity between Neurons and their operations. This will still be Neural Network for data shared and threshold followed for marking differences – Markers, between the two inputs and simplified training. The second Module will be a traditional Neural Network for detection and classification that will track the detected objects. This paper proposed a system that provides the combined image as an output from multiple cameras feed using an untraditional Mathematical and Algorithmic Approach.