Algorithm Research Of Pedestrian Detection Based On Vision Sensor

This paper presents an approach for detecting pedestrians from moving background which is based on compound feature which combined with motion and static feature. It’s difficult to discriminate human body from moving background. We improve a second-order gradient optical flow algorithm Nagel-a. By modifying the Laplace operator estimates of the velocity component, the velocity components in previous iteration are introduced in iterative formula. Nagel-a improves the sensitivity to weak motion of optical flow and enrich the inner repeatability of MBH (Motion boundary histograms) and IMH (Internal motion histograms) motion feature based on the flow. We train a linear SVM (Support vector machine) classifier using features made from pedestrian sample. A fixed window slide over image and classify results are optimized Using Mean shift algorithm. The accuracy is 98% on test of 1,093 group images which better than some previous methods.We present a new method for obstacles and pedestrian detection with lower computation complexity and higher detection performance on small target. By calculating stereo disparity according to the road Parameters, our system efficiently detects objects above the ground. Since the surface parallax mapping algorithm has low computational complexity and calculations can be completed during initialization phase, the efficiency of the detection algorithm is greatly improved. Experiment on over ten thousand images captured in urban area demonstrates our method’s effectiveness on road conditions like asphalt, lawn way, dirt road, slope, night and rainy road.We also propose a multiscale compatible pedestrian detector. Detection speed is improved by avoiding size adjustment of input. Experiment on three public pedestrian database and our HENU database shows that our detector achieves better results than state-of-the-art detection quality at equal speed, especially on small scale pedestrian.An effective approach for pedestrian detecting is presented in this paper considering the characteristics of infrared images and human shapes. A method based on statistics histogram of contrast is used to calculate saliency maps (SM) of infrared images. Attention points (AP) extracted from SM are used to decide segmentation threshold. The hierarchical part-template match tree is built combined with prior probabilities and varieties of human posture. After matching template-tree with segmented results, pedestrians are presented. Experiment results of the approach are compared with other methods on OTCBVS infrared pedestrian datasets. The results show that the proposed algorithm has the highest precision with less time.