Multiple vehicle tracking and classification system with a convolutional neural network

This paper proposes a traffic monitoring system that detects, tracks, and classifies multiple vehicles on the road in real time using various digital image processing techniques and the process of machine learning based on a convolutional neural network (CNN). With this system, a video camera is installed on the road, and calibration is used to obtain the projection equation of the actual road on the image plane. Several image processing techniques, such as background modeling, background extraction, edge detection, and object tracking, are used to develop and implement a prototype system. The proposed system also uses a transfer learning process that is more efficient than starting CNN from scratch. This maximizes training efficiency and increases prediction accuracy in vehicle classification. Preliminary experimental results demonstrate that multiple vehicle tracking and classification are possible while calculating vehicle speed. The ultimate goal of this study is to develop a single digital video camera system with embedded machine learning process that can monitor and distinguish multiple vehicles simultaneously in multiple lanes.