Video-Based Vehicle Speed Estimation Using Speed Measurement Metrics

Abstract

Camera system is widely used as a road traffic monitoring nowadays but if the system is used as a speed camera, an additional speed sensor is required. In this work, we demonstrate a novel method to estimate speed of vehicle in the traffic video without using the additional sensor. We implement two speed measurement models which are measuring traveling distance of the vehicle in a given unit of time and measuring traveling time in a given unit of distance. To get parameters of the models, we define four virtual intrusion lines on road in the camera view. Then, YOLOv3, DeepSORT, GoodFeatureToTrack, and Pyramidal Lucas-Kanade optical flow algorithm are implemented together to detect and track the target vehicle while moving in the camera view. From the tracking data, pixel displacement between two consecutive frames (before and after the vehicle crossing the line) is measured as Crossing distance. The number of frames that the vehicle uses while moving from the first line to the other lines is measured as Traveling time. These two parameters at each intrusion line are used as speed measurement metrics. Solution of the metrics are solved by using tracking data of 20 vehicles at 9 different ground truth speeds measured by a laser speed gun. Then, the metrics are used to estimate speed of 813 vehicles. Our best accuracy is with MAE of 3.38 and RMSE of 4.69 km/h when comparing to their ground truth speed. The same dataset are tested on a Multilayer Perceptron Neural Network model. It can reach accuracy with MAE of 3.07 km/h (RMSE 3.98 km/h).

Keywords

• Vehicle speed estimation
• DeepSORT
• YOLOv3
• GoodFeatureToTrack
• Lucas-Kanade
• optical flow