Shared-use motion vector algorithm for moving objects detection for automobiles

Abstract

This paper presents a new motion vector based moving object detection system for use in Advanced Driver Assistance Systems. By the shared-use of motion vectors with a video encoder, the computational cost for motion estimation using optical flow method can be saved, making the system more cost effective. Moving objects are detected by evaluating the planar parallax residuals of macroblocks in the video. The proposed algorithm involves newly proposed "APD" constraints on hypothesis generation and template matching in hypothesis verification. Test results show that moving objects can be detected effectively in scenarios that pose danger to the ego vehicle.