Moving Object Detection through Efficient Detection and Clustering of Reliable Singular Points.

This paper presents a method to detect vehicles from a moving camera. The detection component involves a cascade of modules. First, motion estimation of singular points in video sequences is used to detect moving vehicle. Finding the image points is the selection of singular points. We impose restrictions about the gradient magnitude and the cornerness. A sparse motion vector field outlines the image motion. We obtain excellent results in real sequences in China. At last, vehicles are detected by clustering of singular points.