Convolution Neural Network Based Real Time Moving Object Detection

Abstract

Moving object detection is the problem of estimating the trajectory of an object in the image plane as it moves around a scene. Here, the focus of work is to develop a framework to detect moving objects, dealing with classification problem in order to recognize the objects unambiguously and then re-identify the object. In this work, frames are extracted from the video captured from static camera. Algorithm is designed to find out a moving object coming in front of camera. A convolution neural network classifier is also designed to classify the object. A background subtraction and filtering (median and bilateral) is done in preprocessing part to train and compile the classifier. This project identify the moving object in different scenario and also able to re-identify the object if it come in front camera again. The results are obtained for LASIESTA moving object image sequences and a recorded dataset. The proposed work shown improved performance over the existing object classification techniques.