Multiple moving object Detection and Classification using block based feature extraction algorithm and K-NN Classifier

Abstract

Surveillance system is used to check the nature or functioning of objects in video scene. It also helps to detect, track and classify multiple moving vehicles like car, bike, bus and truck using feature extraction technique. In video surveillance detection and classification is a very big challenge based on the shape of multiple moving objects. An algorithm or framework is designed for detection and classification of multiple moving objects such as vehicles in this research. Background subtraction is used for detection of multiple moving objects like vehicles using Gaussian mixture model (MOG). Block-based feature extraction approach is applied to get different types of features in video frames. K-nearest neighbor classifier is also used in both feature extraction and classification approach based on different features which include size, descriptors of shape in this research. An algorithm discussed in this paper gives more accurate value (result) rather than multiple virtual detection line (MVDL). This algorithm also improves the accuracy value and reduces the detection (counting) and classification error for multiple moving vehicles in video frames. This approach is also useful for removing occlusion and variable illumination problems in video scene.