Research And Implementation Of Pedestrian Detection Algorithm In Video Stream

Pedestrian detection is the most active and challenging field of computer vision.Pedestrian detection is a branch of target detection,the purpose of which is obtaining the accurate position of pedestrian from consecutive video frames.Compared to pedestrian detection technology under the static background,pedestrian detection technology under the dynamic background is vulnerable to influenced by illumination.So,pedestrian detection is comparatively difficult.The paper studied pedestrian detection algorithm and implemented the function of pedestrian detection based on the Linux2440 platform.The paper adopted a pedestrian detection method combined head feature recognition algorithm and HOG&SVM algorithm on matlab2014a platform.In this paper,the pedestrian detection is mainly divided into extraction of region of interest(ROI)and pedestrian recognition.The existence of noise is inevitable in the video image of ROI extraction algorithm.The first step is to apply average filter and median filter to filter out the noise in the image,which can improve the signal-to-noise ratio.The second step is use inter-frame difference algorithm to obtain moving object and use mathematical morphology method to compensate hollow produced by inter-frame difference algorithm.In order to obtain the properties of connected component area,the third step is to label the connected component tags for moving object.Pedestrian recognition algorithm apply the geometrical characteristics of pedestrians to separate pedestrians from video image,and then use head feature to obtain pedestrian targets in video image.Finally,extracting HOG features of the pedestrian targets obtained by the previous two steps,which is sent to the SVM classifier to recognize pedestrian,so as to improve the accuracy of the pedestrian detection.The realization of algorithm based on hardware platform contains several steps.The first step is to build Linux platform.The second step is to write code of image acquisition based on V4L2 and transmission based on TCP/IP and pedestrian detection on the QT in Ubuntu.The final step is to use cross compiler to generate executable code and adjust the system on hardware platform.The result of hardware tests shows that the system effectively realizes the pedestrian detection,and the system has high detection rate,good real-time detecting and robust character.So it has a wide application prospect.