Design and Implementation of an Embedded Pedestrian Detector

Abstract

Pedestrian detection in complex scenes is a key challenge in traffic detection by video. A newly designed, embedded pedestrian detector based on the kernel TI's DSP chip (TMS320DM642) is proposed. The detector adopts the Hausdorff distance template-matching algorithm for pedestrian detection and tracking. The framework of the algorithm and its DSP optimization procedure is described in detail. The results of a real crossing experiment indicate that the detector is able to rapidly detect pedestrians in a complex traffic scene with 95% precision. Video-based pedestrian detection is a valuable practice among fields of intelligent traffic control, intelligent community monitoring, and supermarket passenger flow estimation. With advances in image processing technology and cost-effective improvement of hardware, pedestrian detection has been widely used in many areas. In recent years, with the development of embedded technology, especially the maturation of the digital signal processors (DSP), embedded pedestrian detection has been improved in both hardware and software. The unique stability of DSP, repeatability of function, superior data processing capability, and especially its programmable and easily implemented adaptive processing, have tremendously developed the real-time image processing field. In comparison with PC machine that is vulnerable to environmental interference, the ability of DSP chip for a particular processing task has considerable advantages for intelligent transportation systems. The pedestrian as an object is a complex pattern influenced by many factors, and finding an effective way to extract the common features of the pedestrian in order to describe the pedestrian model is a key for pedestrian detection. Through combining the characteristics of pedestrian detection and DSP, a DSP-based embedded pedestrian detector which includes the Hausdorff distance template matching algorithm for real-time detection and pedestrian tracking system is constructed through the development platform of TI's DSP chip TMS320DM642 for multimedia applications.