Pedestrian and car traffic analysis by computer vision – classification development

Abstract

This project presents a technique for classifying the regions for moving cars and walking pedestrians on a road. A static camera is required to acquire video sequences. The data obtained by this classification algorithm is useful for traffic analysis and behaviors monitor of cars and pedestrians. In this project, optical flow is used to find the velocity vector field which indicates the moving directions of the pixels. Then, Gaussian mixture models (GMMs) are obtained by applying the regularized Expectation-Maximization (EM) algorithm. The GMM represents a continuous probability density function of the direction of the vector for a pixel. Finally, delta rule training algorithm is used to train the GMM in order to classify it. For the classification performance, the classification algorithm developed can successfully classify the region for moving cars and walking pedestrians and the background can be extracted.