Font Size: a A A

Research Of Abnormal Detection In Crowd Scenes Based On Sparse Analysis

Posted on:2018-02-13Degree:MasterType:Thesis
Country:ChinaCandidate:F X DongFull Text:PDF
GTID:2348330536479532Subject:Signal and Information Processing
Abstract/Summary:PDF Full Text Request
In recent years,with the continuous improvement of public awareness of public safety,intelligent video surveillance system has developed rapidly.Through the effective video analysis intelligent video surveillance can recognize and judge the target or the event in the scene automatically.Video abnormal detection is an important research topic in the field of intelligent video surveillance,and has attracted more and more researchers' attention.In this thesis,we study the crowded abnormal detection technology of crowded scenes based on sparse representation.The main contributions and results are as follows:1?Video feature extraction of crowd abnormal detection based on sparse representation has been researched,two types of fast feature extraction methods based on image binarization are proposed.One is the global binarization HOG feature extraction method(GB-HOG).At first the difference frame between the current frame and the previous frame is obtained by using the frame difference method,and then the difference image is binarized with the OTSU thresholding,at last the HOG feature is extracted directly on the binarized image;Anther is the region original HOG feature extraction method(RO-HOG).At first the difference frame between the current frame and the first few frames is obtained by using the frame difference method,and then the difference image is binarized with the OTSU thresholding,the intersection of non-zero region on the binarized image is detected to obtain the main active area,last the HOG feature is extracted from the main active area.Experiments show that by using these two types of feature extraction methods in crowded abnormal detection based on sparse representation,the speed of abnormal detection can be improved significantly.2?Dictionary model selection of crowd abnormal detection based on sparse representation has been researched,crowded anomaly detection based on adaptive weighted double dictionaries is proposed.Our method uses the GB-HOG and RO-HOG feature extraction methods proposed above to extract features for double dictionary learning,and then the test sample features are sparse represented on the double dictionary to compute the sparse reconstruction costs.At this time,based on the uniformity of the scene calculated,the Gaussian function estimator is used to estimate the weights of the two parts of sparse reconstruction cost,the weights of the two parts are used to weighted sum of the two parts of sparse reconstruction cost,at last the final sparse reconstruction cost is obtained to judge anomaly events.Experiments show that through the realization of double dictionary,the accuracy of abnormal detection can be improved.3?Dictionary update of crowd abnormal detection based on sparse representation has been researched,crowded anomaly detection based on dictionary incremental updating is proposed.In this method,by setting a criteria to determine the new features for the dictionary incremental update,then the new video features are adopted for the dictionary relearning according to a certain frequency,so that the dictionary can represent the normal sample feature more accurate.Experiments show that by using the dictionary updation,the robustness of abnormal detection can be improved.This thesis analyzes the experiments and performance of the above algorithm on different video sequences.The experimental results show that by using these two types of feature extraction methods in crowded abnormal detection based on sparse representation,the speed of abnormal detection can be improved significantly.The proposed crowded abnormal detection algorithm based on the adaptive weighted double dictionaries can adjust their own abnormal detection algorithm according to the different conditions of the scene to improve the accuracy of the crowded abnormal detection algorithm.The proposed crowded abnormal detection algorithm based on the dictionary increment update can update the dictionary according to the scene changes,the robustness of the crowded abnormal detection algorithm is further improved.At the end of this thesis,it summarizes the work of the thesis,and makes further discussion on the possible improvement in the future.
Keywords/Search Tags:crowd abnormal detection, sparse representation, fast feature extract, adaptive double dictionary, dictionary increment update
PDF Full Text Request
Related items