Biologically Inspired Composite Vision System for Multiple Depth-of-field Vehicle Tracking and Speed Detection Lin Lin, Ramesh Bharath * , and Xiang Cheng Department of Electrical and Computer Engineering, National University of Singapore, Singapore 117576 Abstract. This paper presents a new vision-based traffic monitoring system, which is inspired by the visual structure found in raptors, to provide multiple depth-of-field vision information for vehicle tracking and speed detection. The novelty of this design is the usage of multiple depth-of-field information for tracking expressway vehicles over a longer range, and thus provide accurate speed information for overspeed vehicle detection. A novel speed calculation algorithm was designed for the com- posite vision information acquired by the system. The calculated speed of the vehicles was found to conform with the real-world driving speed. 1 Introduction Object tracking has been one of the most attractive topics in computer vision, and it has various practical applications so far, such as human-computer inter- action [1, 2], video surveillance [3–5], vehicle navigation [6], traffic monitoring and control [7–9], and motion analysis [10]. In this paper, the focus is on traffic monitoring in expressways for automatic overspeed vehicle detection. In the past few years, various vision-based methods have been designed to solve the traffic monitoring problems using a single [11–14] or stereo camera [15,16]; however, the performance of the above-mentioned systems is limited by a small tracking range due to the fixed depth-of-field of the cameras [17, 18]. These systems per- form vehicle tracking near the installed location, and hence they can only track and calculate vehicle speed within a small distance. Therefore, these systems are better suited for traffic monitoring situations such as congestion control and intersection monitoring. In practice, a long tracking range is crucial when high- speed vehicle monitoring is needed. In contrast, the traditional speed detection approaches using sensors such as LIDAR/RADAR have several drawbacks [19,20]. These approaches gener- ally work in this way: the sensors detect the presence of a possible overspeeding vehicle and trigger a camera to capture the image of the overspeeding vehicle. However, with a large amount of vehicles present in the scene, the detector might know one of the vehicles is overspeeding but not be able to single it out. Further- more, the interference caused by big vehicles leads to unreliable results for speed * Corresponding author.