Vision-Based Vehicle Speed Estimation

This paper presents a system to measure the speed of vehicles on roads. Traffic law violations are controlled by detecting over speeding vehicles. Vehicles in a video are detected, tracked and their speed is estimated. Speed estimation is performed based on camera input without any extra sensors making it cost effective. Fast Retina Key-point (FREAK) and Features from Accelerated Segment Test (FAST) algorithms are used for feature extraction. FAST and FREAK provide results rapidly and are helpful in real time applications. Time calculation for speed estimation is independent of the processor being used. Voting based classifier is used for detecting vehicles. Seven different classifiers are used in it. Random Forest provided the highest accuracy of 88.4% and F1 score of 88.5%. Proposed system provided speed estimation with an approximate error of 2 km/hr. The average percentage error obtained for the estimated speed is 9.22%.