Font Size: a A A

The Research On Pedestrian Detection Algorithm Based On Feature And Depth Map Fusion

Posted on:2021-02-07Degree:MasterType:Thesis
Country:ChinaCandidate:Y WuFull Text:PDF
GTID:2428330623467320Subject:Electronic and communication engineering
Abstract/Summary:PDF Full Text Request
As an important research hotspot in the field of computer vision,pedestria n detection is mainly used to detect pedestrians from images or video data thr ough image processing,pattern recognition and other computer vision technolog ies,and accurately track to the pedestrian target to determine the position of p edestrians.Pedestrian detection is not only applied to simple pedestrian detection,it is closely related to intelligent monitoring,intelligent traffic,human behavio r analysis and aerial image,and has high value in scientific research and busin ess.At present,the basic features of pedestrian detection are HOG,LBP and Haar-like.However,single features have the characteristics of slow speed,diffi cult calculation and poor real-time performance.Therefore,there are some short comings in pedestrian detection.In this paper,under the original single feature pedestrian detection,the multi-feature pedestrian detection,the Faster RCNN method and the fusion depth map pedestrian detection are studied.The main c ontents are as follows:1.Aiming at the shortcomings of pedestrian detection based on single feat ure,the multi-feature fusion is further analyzed.Through the feature extraction and the training of the classifier,the representation ability of pedestrians is enh anced.At the same time,the detection method of Faster RCNN based on deep learning is also studied,which improves the performance of pedestrian detecti on.2.Aiming at the problem that hog-lbp fusion feature pedestrian detection i s greatly affected by the environment,this paper proposes a pedestrian detectio n method based on feature and depth map fusion.By extracting the HOG featur e of the color graph and the CLBC feature of the depth graph,the weighted f usion method is adopted for fusion.Meanwhile,the kernel function is used to cross the kernel to improve the classifier,so as to obtain more depth informati on from the picture and improve the detection accuracy.3.Based on VS12 and Opencv,a pedestrian detection system is designed,which is composed of three modules: background modeling,merging and delet ing of connecting areas,and pedestrian detection,realizing the detection of mul tiple pedestrians in a complex environment...
Keywords/Search Tags:pedestrian detection, bottom features, depth map, weighted fusion, background modeling
PDF Full Text Request
Related items