Vehicle tracking and speed estimation using view-independent traffic cameras

Abstract

The traditional method's speed estimation accuracy depends on environmental changes and the new CNN technique accurately estimates speed only for quality video but input lacks, which leads to overfitting. Hence proposed method uses the Kalman filter and Hungarian algorithm to track detected vehicles for estimating the speed. First, vehicles are detected using the k-nearest neighbour (kNN) classifier-based background subtraction technique. Then logistic regression value is mapped to a binary value called decision threshold which accurately detects the vehicles. Secondly, each detected vehicle is tracked by a Kalman filter, then the Hungarian algorithm connects all the predictions and produces the tracking results. Finally, the vehicle speed is estimated by measuring the distance between the detected vehicle and the virtual line tracked vehicles on the ground plane. It also stores over-speeding vehicle images with speed. The proposed technique is verified with the BMVC standard database which shows a maximum error of ±2 kmph.