DEVELOPMENT AND OPTIMIZATION OF ALGORITHM FOR VEHICLE TRACKING ОN VIDEO USING LUCAS–KANADE METHOD WITH PYRAMIDS

Abstract

The paper considers the task of traffic flow analysis when a stationary video sensor observes a road section. An algorithm for tracking vehicles moving along road lanes is presented. This algorithm includes the stage of extracting feature points of the object part located in a bounded area of the image. Object area extraction is based on preliminary segmentation results and information about scene geometry. The second stage of the algorithm uses the Lucas–Kanade method with pyramids to track the vehicle. The positive features of this method for optical flow calculation are considered. Approaches to optimizing computational costs in optical flow calculations are proposed. Information about movement direction and speed of the vehicle is used for this purpose. The developed tracking algorithm has been tested during experimental studies. The studies were carried out using real videos оn which vehicle movement on several road lanes is observed. The results of experimental studies have shown the high performance of the proposed algorithm in comparison with the basic algorithm for moving object detection without further tracking. As a result, the number of false detections and omissions of vehicles decreased. The percentage of classification errors also decreased which indicates an increase in the accuracy of estimating the length and speed of detected vehicles.