Font Size: a A A

The Person Re-identification Algorithm Based On Bag-of-words Model And Pedestrian Attributes

Posted on:2017-06-01Degree:MasterType:Thesis
Country:ChinaCandidate:J Q ShiFull Text:PDF
GTID:2348330488458153Subject:Information management and e-government
Abstract/Summary:PDF Full Text Request
In recent years, with the increasing popularity of video surveillance systems, video analysis technologies play an increasingly pivotal role in public security and criminal probing. Pedestrian retrieval problem without overlap perspective, namely, the person re-identification now becomes an important research topic in security surveillance. Therefore, the establishment of a robust and stable person re-identification algorithm is of great significance in research and application.This paper presents a person re-identification algorithm based on the bag of words model and pedestrians semantic attribute. The main contents are as follows.(1) We propose a person re-identification algorithm based on bag-of-words model and comprehensive sorting of multi-feature matching result. Firstly, to overcome the shortcomings of traditional bag of words model, vertically divide the target proportionally and optimize the calculation method of similarity combined with the spatial structure information of pedestrians and prior knowledge. Secondly, the similarity of the images is calculated according to the feature position. Since the pedestrians images are uniformly proportional divided, the local features extracted are not only coupled with the overall structure information, but also able to be matched by low-dimensional visual words, which thereby can reduce the computational complexity and are more in line with actual application requirements. Besides, on the basis of bag-of-words model, we propose a person re-identification algorithm of comprehensive sorting of multi-feature matching result. Improve the recognition accuracy by weighting each of sample over calculating the position in the other sorted lists. The proposed method has been tested on VIPeR and Market-1501 data set and compared with existing similar methods. Experimental results show that the first recognition accuracy has improved.(2) We propose a person re-identification algorithm with the fusion of semantic attributes feature and underlying visual features. Firstly, feature prediction model is constructed by SVM classifier for acquiring the pedestrian feature characteristics. Secondly, the results is optimized with consideration of the correlation between pedestrian attributes. Finally, we propose a method of similarity calculation based on semantic attributes and metric learning. The proposed method has been tested on VIPeR and GRID data sets. The attributes features make up for the shortcomings of the lack of image semantic information of the underlying visual features, which to some extent, improve the accuracy of person re-identification.In conclusion, this article have completed some meaningful attempt and exploration aiming at the person re-identification. All these results contribute to further enrich and improve the related theory, method and technology in the field and it would have a bright future.
Keywords/Search Tags:person re-identification, bag-of-words, priori knowledge, comprehensive sorting of multi-feature matching result, semantic attributes feature
PDF Full Text Request
Related items