Computer Vision Research Group

Automatic Vehicle Speed Esitmation from CCTV Feeds

Currently radars /laser based systems are used to estimate velocity of vehicles on highways. However these systems require human interaction to detect the speed violations by vehicles. In this project we aim to develop a system which will accurately estimate the velocity of vehicles using CCTV cameras and detect any speed limit violations. Since CCTV cameras are already deployed on freeways for the purpose of traffic monitoring and surveillance. So the proposed system will also reduce the overall cost of velocity estimation system.

CCTV cameras are not calibrated and can be panned tilted and zoomed. Image sequences from these cameras do not preserve geometric properties (length ratios and line parallelism). These images can be rectified to restore these geometric properties. Image rectification requires two vanishing points. The vanishing point in the road direction is extracted automatically by exploiting lane demarcations. One vanishing point is calculated using RANSAC algorithm while the other vanishing point which is orthogonal to the calculated vanishing point is assumed to be at infinity. Thus, the projective distortion of the road surface can be removed allowing affine rectification.

This approach requires known distance between two points 'a' and 'b' in world plane. Moving vehicles are detected by background subtraction algorithm and tracked between these two points using Lucas Kanade Tracker. Vehicle speed can be estimated if the frame rate and distance between two points 'a' and 'b' is known.

Original Image

Rectified Image

Vehicle Detection and Tracking

© 2010 Department of Electrical Engineering, CIIT Lahore  |  Terms and Conditions