Research On Detection Technology Of Pedestrian Flow Based On Intelligent Video Analysis

Intelligent video analysis technology has received widespread attention since its rise and has been widely applied in the field of public safety,city management etc.It relies on the advanced technology of motion detection,feature matching and pattern recognition,and commonly used in video surveillance scheme.As an important application of this technology,people flow detection has extremely extensive application requirements.The traditional infrared sensor flow detection method is difficult to judge pedestrian direction,thus it can't achieve the accurate detection of people flow.People flow detection based on intelligent video analysis can automatically realize the detection of surveillance video by video image processing technology.This technology can not only achieve the number of pedestrians and judge the direction of pedestrians,but also to make timely warning of the mutation,to avoid hidden danger.Therefore,it has important theoretical significance and application value to carry out research on traffic detection technology based on intelligent video analysis.Based on the advantages and disadvantages of the-depth analysis of pedestrian detection method based both on motion analysis and statistical learning,the paper studies how to conduct effective detection of pedestrian in video images according to the distribution characteristics,how to improve the problem of long detection time caused by the sliding window search method.The main work of the thesis and academic achievements includes:1.The paper briefly introduces the research background and application requirements of the people flow detection technology which based on intelligent video analysis.As the key technology of people flow detection technology,the paper introduces the research status of the pedestrian detection both from theory research and practical application and the existing pedestrian detection technology are classified as well as summed up the advantages and disadvantages of each method.2.The paper describes the basic principle of target detection based on motion analysis.Three methods,i.e.,frame subtraction method,optical flow method,and background subtraction method,were used to detect target of a test video.Then the paper summarized the scene,complex detection,detection accuracy rate of each method according to analysis of detection results.3.There are two shortcomings in the previous pedestrian methods based on motion analysis.One is the “empty” phenomenon,and the other is weak real-time performance owing to the addition of the pedestrian classifier.To solve these problems,a new pedestrian detection method based on statistical learning is proposed.Firstly,an improved HOG-LBP feature set is proposed to address the high dimensionality and the redundant information of HOG and LBP features.The optimal HOG feature set is selected via the statistical average of the original hog features according to individual optimal feature combination and Bhattacharyya distance discriminability criterion,and then is combined with original LBP features to get the improved HOG-LBP features.Secondly,the sample characteristics are trained by support vector machine to get a classifier,which used for pedestrian detection.Experimental results show that the proposed method obviously shortens the time of feature extraction,but still keep the detection accuracy.4.To address the problem of the low detection speed caused by the classical sliding window search method over non-target window detection in the traditional pedestrian detection process,a new pedestrian detection method composed of BING prediction and DPM recognition is proposed.Firstly,the suspicious target candidate regions are obtained quickly via BING prediction,and are selected according to the prior knowledge of pedestrian and the scores of candidate regions.Then,the filtered candidate regions are identified by the pedestrian classifiers with parameters optimized.Finally,the recognition results are post-treated through non maximum suppression algorithm to determine the accurate location of pedestrian target.Experimental results show that the detection rate of the proposed method is improved nearly 4 times over the traditional pedestrian detection,thereby realizing effective detection of the pedestrian target.5.A pedestrian flow detection system is designed based on MFC to mainly achieve the accurate counting and real-time display of pedestrian flow information.The realization process is composed of three parts: pedestrian detection,pedestrian tracking and pedestrian counting.Pedestrian detection method is based on the combination of BING prediction and DPM recognition.Pedestrian tracking method is based on mean-shift combined with Kalman.Pedestrian counting method is based on trip line counting.Finally,the paper summarizes and prospects the development and research of pedestrian flow detection technology based on intelligent video analysis.