A New Robust Multiple Pedestrians Detection and Tracking System

As part of DOT's Intelligent Vehicle Intiative program, the FHWA is developing advanced vehicle safety and driver information systems. Vision-based pedestrian detection in outdoor scenes remains an open challenge. People dress in very different colors that sometimes blend with the background. They wear hats or carry bags, and stand, walk and change direction unpredictably. The background is various, containing buildings, moving or parked cars, street signs, etc. Moreover, sudden changes of background are inevitable in vison systems mounted on a moving vehicle. Aimed at these difficulties, in Phase I, we propose to develop a new real-time pedestrians detection and tracking system. In particular, we propose to develop the innovative technologies including: 1) adaptive optimal wavelet thresholding for disparity image denoising, which will overcome the problems associated with morphological opeators when the blobs resulted from noises are large or connected with large objects; 2) real-time detection of pedestrians based on the estimation of local 3D features used to approximate the legs, upper body and head of a pedestrian; and 3) pedestrian discrimination based on statistical hypothesis tests for pedestrians. Our system does not require any prior templates which are often difficult to select for the real-world applications.