Multiple Cue Based Vehicle Detection and Tracking for Road Safety
Michael Wan, Sarath Kodagoda, SA Sehestedt · UTS ePRESS (University of Technology Sydney) · 2007
With the rise in accident related fatalities on roads, the researchers around the world are looking for solutions including integrating intelligence to vehicles. One cruicial aspects of it is the robust detection and tracking of other vehicles in the visinity. In this paper, we have proposed a probabilistic way of incorporation of several visual cues in vehicle detection and a particle filter based tracking strategy. Visual cues used are, lane markings, symmetry, entropy and shadows. Combination of visual cues provided us with robust results when compared with their individual counterparts. The definition of a region of interest lowers the computational requirements with improved robustness. Experimental results of the algorithm in Sydney urban areas are presented