Pedestrian Detection Algorithm Research Based On DM642

The pedestrian is one of the important participants of urban road traffic, so to carry out the research of pedestrian detection technology has great significance for the development of the urban intelligent transportation system. Pedestrian is a non-rigid target, because of the complexity of its scenes and posture, shape changing, makes pedestrians video detection, especially the video detection based on embedded devices has been a difficult challenge in the field of computer vision research. As the intelligent transportation system increasingly intense demand of the miniaturization, low power consumption, low cost, stable and reliable product, and carry out the research of embedded pedestrian detection technology has the important practical significance and application value. This thesis is based on TI company’s digital signal processor (DSP) TMS320DM642platform, researched the pedestrian detection algorithm, and implementation the pedestrian detection system. The main content of the thesis and the research results include as follows:(1) Algorithm research and design. This paper based on the principle of Hausdorff similarity measurement, designed a pedestrian’s head and shoulders edge character template, then combining motion area detection, template matching algorithm and kalman filter tracking algorithm, proposed the rapid pedestrian detection algorithm based on Hausdorff distance template matching. The algorithm combining detection speed and detection accuracy, easy to transplant in embedded platform. The experimental results show that this algorithm detection speed can reach to27frames per second on PC, and detection rate can reach95%, detection error rate is only3%, meet the demand of actual use.(2) Algorithm implementation and optimization. This paper implementation the rapid pedestrian detection algorithm based on Hausdorff distance template matching on DM642platform, and through the use of inline functions, loops unrolling, rearrangement of software pipelining, using a ping-pong mechanism to transmit data and other means to optimize the algorithm and program. Actual intersection test results show that the detection speed of the rapid pedestrian detection algorithm based on Hausdorff distance template matching can achieve9frames per second on DM642embedded platform, it reaches the real-time detection, and the pedestrian detection rate not less than94%, the detection error rate is not higher than3%.Experimental results show that the rapid pedestrian detection algorithm based on Hausdorff distance template matching on DM642embedded platform this thesis designed and realized can carry on the real-time detect and track of pedestrians, and guarantee the accuracy of detection in this premise. This pedestrian detection algorithm perform well on the balance between detection speed and detection rate, and lay a certain foundation for further research and application of the embedded pedestrian detection algorithm.