Biologically-inspired multi-object tracking algorithm applied to traffic monitoring

Various sensing modalities are presently used or conceptualized for use in roadway management projects. These include, among others, bind-type (utilizing inductive loops, reflected light, etc.), acoustical, and radar sensors. However, due to lower cost; flexibility in installation, maintenance and use; high information content; lack of emissions; and wide sensing area, some researchers have argued that the use of video cameras and computer vision techniques offer the best choice. Nonetheless, due to various issues - including lighting condition sensitivity and computational complexity - even the latter approach can be less than optimal. One possible reason that a computer vision scheme can become overly complex and sensitive to lighting conditions is that too much local information needs to be captured/analyzed by the algorithm. Utilizing a patent-pending approach, this paper discusses a just-enough-smart, largely-global technique based on a simplified version of the fly's eye that can address these latter issues. This approach can easily generate vehicular flow data - such as entrance, exit and throughput rates - or serve as a front-end process for yielding pointers to vehicles to assist in the analysis activities of other algorithms. Results from the application of this approach to the tracking of multiple moving vehicles or people in real scenes with fixed and moving cameras are also presented.