Real-Time Speed Estimation of Vehicles from Uncalibrated View-Independent Traffic Cameras

This paper presents a comprehensive solution for implementing a processing module on traffic cameras that is capable of tracking every vehicle in the camera frame and estimating its speed in real-time. A hybrid framework for multiple vehicle tracking is employed that utilizes Kalman filter and Hungarian Algorithm to resolve occlusions. A speed estimation methodology is described that is robust enough to work with camera feed from any angle without calibration and camera mounted at a minimum height of 7m. The system has been tested on computer-generated sequences as well as real-life scenarios and speed estimates have been obtained with maximum error of less than 3kmph.