Vehicle Speed Estimation with 3D Bounding Boxes

Abstract

Speed Estimation makes use of computer vision and machine learning algorithms in order to detect, keep track, and estimate speed of vehicle under surveillance. Speed estimation can be integrated with other traffic analysis solutions like vehicle counting, classification and license plate recognition system. My approach uses the vanishing points (VP) geometry and scene scale in the scene to calculate a perspective transformation matrix. This transformation makes us able to simplify the task of detecting 3D bounding boxes to the task of detecting 2D bounding boxes with one extra parameter using an efficient object detector. Additionally, I have proposed algorithm of automatic video mask generation which eliminates this manual step from pipeline. This improved algorithm has improved speed measurement accuracy by reducing mean speed measurement error by 5% and the median speed error to 3%.