Fast Pedestrian Detection Method Based On BING And C4

In the field of computer vision,whether it is the image classification,or the target recognition,the traditional method uses a trained classifier to detect the sliding window of the input image.A large number of windows will be generated during the sliding process,and we need to detect the window with classifier,so as to achieve the purpose of detection and recognition.Therefore,the key to improve the speed of detection algorithm is to reduce the number of detection windows.To solve this problem,a fast pedestrian detection algorithm based on BING and C4 is proposed by studying the physical property method and traditional pedestrian detection method.Through the analysis of the detection process of traditional methods,it is known that the sliding window traversing the image is an important factor affecting the detection speed,so a lot of researchers have started the study of the area of interest in the extraction of images,that is,quasi physical sampling.In 2014,Cheng M.M.and others published a CVPR based BING algorithm.The BING algorithm is based on the human visual system.It is like people's eyes can quickly and accurately focus attention on more obvious visual objects in the face of complex scenes,and then the object is carefully identified.Similarly,BING is to find objects contained in an image,ignore the background and interference in images,and quickly locate areas of interest.The core of the algorithm is that if the object in the image is zoomed to 8*8,there will be obvious closed edge features.In view of the different probability of pedestrians in different size images,a second level classification model is added to correct the results of different sizes to prevent missed detection.C4 algorithm is a fast and accurate pedestrian detection algorithm proposed by Wu JX et al.The algorithm uses CENTRIST features based on pedestrian contour information,and then uses cascade classifier to detect input images.Compared with HOG+SVM,C4 algorithm has more advantages in detection speed and accuracy.First,we need to build image Pyramid by scaling.Then we build Sobel image with Sobel operator,and then compute the CT of each pixel to CT image.In the detection stage,linear SVM is used to detect pedestrians.By constructing auxiliary images and integral images,we can get the classification results of linear SVM conveniently.Because the accuracy of linear SVM classification is low,we need further detection using HIK SVM.The complete pedestrian detection method based on BING and C4 takes the BING method as the preprocessing stage of pedestrian detection,and gives the object suggestion window to the accurate pedestrian detection method C4 for accurate detection.In view of the application scene of the detection algorithm on ADAS,we carry out the algorithm on the TX1 platform,and use the INRIA database and the image library data taken by oneself to carry out the experiment,which proves that the algorithm meets the requirements of the detection speed and accuracy.