Realtime On-Road Vehicle Detection

By Jaesik Choi, Dan Parente and Eyal Amir

Abstract

The autonomous vehicle system is a demanding application for our daily life. The vehicle requires on-road vehicle detection algorithms. Given the sequence of images, the algorithms need to find on-road vehicles in realtime. Basically there are two types of on-road vehicle (cars traveling in the opposite direction and cars traveling in the same direction). Due to the distinct features of two type of vehicle, different approaches are necessary to detect each direction. Here, we suggest the optical flow to detect the coming traffics because the coming traffics represent distinct motion. We use Haar-like feature detection for the traffics in the same direction because the traffics represent relatively stable shape (car rear) and little motion. We verify the detected region with estimating 3D geometry. If the detectors fail to find the vehicles, we interpolate the region from the previous frames. Then, the detected vehicles are projected into the 3D space. Our system detects vehicles with high accuracy and realtime for 11 frames/sec.

Related Papers

Updated on Feb. 27, 2007