Fast Pedestrian Detection Method Based On CBING And ChnFtrs

In computer vision,whether it is on the image classification,or target identification and tracking,the traditional methods need to train specific features classifier according to the features,and the input image to do sliding window scan.This will produce a large number of suggested windows,through the classifier to assess the window in order to achieve the purpose of detection and identification.Reducing the number of recommendations for the classifier evaluation windows is particularly important for improving the speed of target detection and identification.Aiming at this problem,this paper has carried out an in-depth study on the similarity in the field of saliency and the traditional pedestrian identification,and proposed a fast pedestrian detection method based on CBING and Chn Ftrs.The main three contributions of this paper:1.Proposed an improved object suggestion method,CBING,is propose,which can effectively propose a suggested window that may contain objects and reduce the next stage of accurate pedestrian detection area.In the CBING method,combined with the idea of BING algorithm,find the edge of the object and non-object Candy edge characteristics.In order to combine the NG features of the original method,the training method of cascaded two-stage SVM is proposed,which effectively combines the advantages of the two features and achieves the improvement of performance.In the stage of positive and negative sample extraction,a negative sample extraction algorithm is proposed,which can effectively improve the difference between positive and negative samples and reduce the redundancy of negative samples.At the same time,we focus on supplementing pedestrian samples.2.Improved Chn Ftrs algorithm.Haar-like can be used to reflect the characteristics of pedestrian junctions and integrate Haar-like features into Chn Ftrs integral channel.Experiments show that the improvement can improve the robustness of pedestrian detection algorithm effectively.3.Based on CBING and improved Chn Ftrs,a complete pedestrian detection method is proposed.The CBING method is used as a pre-processing stage for pedestrian identification,and the resulting object recommendation windows is accurately identified as a possible pedestrian area for the improved Chn Ftrs + Adaboost and tested under the INRIA database and the self-built pedestrian database.The experimental results show that the proposed window proposed by the CBING method has a significant improvement in the DR(detection rate)and MABO(detection accuracy)relative to the BING method and the state of art detection methods.Based on the CBING and improved Chn Ftrs combination of pedestrian detection methods,to ensure Chn Ftrs pedestrian recognition performance,while greatly improve the pedestrian detection speed.