Vehicle recognition and its trajectory registration on the image sequence using deep convolutional neural network

The article shows the methods of vehicle recognition on the image sequence and its trajectory registration. As a recognition algorithm authors used Viola-Jones method with optical flow filter and the deep convolutional neural network in combination with sliding window technique for vehicle detection task. Also authors analyze approaches to registration of detected vehicle trajectories on image sequence based on its linear and angular velocities and Kalman filter. The efficiency of vehicle detection is shown in terms of the precision and recall of recognition. Quality of vehicle registration on the image sequence is estimated by the standard deviation of results from sample values. The article also shows usage prospects of proposed algorithms as a part of driver assistance system and unmanned vehicle control system.

I. Introduction

Machine vision methods are actively researched in direction of development of driver assistance systems and unmanned vehicles to ensure trouble-free driving and road safety. Vehicle recognition and tracking its trajectory on image sequence is one of such methods. On the first hand, its implementation involves the solution of the vehicle classification and detection problem and, on another hand, registration and prediction of their trajectory on the basis of the video stream obtained from moving cameras. A number of approaches to solve both problems have been developed in the world.