Research On Road Pedestrian Recognition Algorithm Based On DSP For Driver Assistance System On Vehicle

Pedestrian recognition, a core issue of the research area of computer vision, has awide range of application in many areas, such as robot, surveillance system andautomatic security system, and is the research hotpot of the auxiliary driving,intelligent vehicle and transport. With the development of computer technology、electronic technology、machine vision、 pattern recognition and image processing、artificial intelligence technology, pedestrian recognition has been developed greatly inthe multi-interdisciplinary.Although the theoretical system of pedestrian recognition is more and morematured, the existing systems are all relied on PC which makes them not suitable forapplication in vehicle assistance driving of security field with higher demand. Now,the application of existing pedestrian recognition system on vehicle is only installedon the vehicle with higher configurations. With the progress of DSP technology, theembedded pedestrian recognition system, which sets the pedestrian acquisition infront of the vehicle, processing and recognition as a whole, has a broad research valueand business prospective.To achieve the pedestrian recognition system based on DSP in vehicle assistance driving, we must first achieve that the acquisition and display of the pedestrian videoand image and the pedestrian recognition algorithm are embedded in DSP successfully.This paper focuses on the two issues and conduct the follow works.(1)Acquisition and display of the pedestrian image. The CCD camera, which isinstalled on the windshield in front of the vehicle and has the same height with therear view mirror, is selected to acquired the pedestrian video image. The collectedvideo image is decoded by decoder chip TVP5150, then the decoded data is storedinto SDRAM through EDMA, which is encoded by encoder chip SAA7175afterimage processing and displayed by monitor.(2)Design of the DSP/BIOS application program frame based on RF5. This paperdesigns three tasks on the basis of embedded by encoder data is stored into SDRAMthrough EDMA, which is encoded by encoder chip SAA7175after image processingand displayed by monitor.(3)Achievement of the pedestrian recognition on DSP.In embedded DSP system,because of the restriction of data format and storage space, the execution andprocessing speed of algorithm is slower than PC, making the algorithm must besimple and accurate.The first step of the algorithm in this paper is to acquire theregion of interest of pedestrian and the specific method is binarization processing utilizing adaptive maximum between-class variance method, and then filteringprocessing. Finally, the lane line is extracted by improved HOUGH change and theregion of interest of pedestrian is collected by filtering the image information exceptthe lame line. The second step is pedestrian recognition. First, mark the candidatepedestrian area from the region of interest according to the shape feature of pedestrian,then calculate the texture features of the acquired candidate region and classify themby utilizing the trained classifier to determine whether it is a pedestrian.(4)The hardware and software of this system are debugged and the real-time andCPU processing power are analyzed. The pedestrian recognition program is optimizedpartly to improve the utilization rate of program and this word obtains the effect ofreal-time operation, which illustrates that the road pedestrian recognition algorithmutilized in this paper has a high reliability.