Pairwise Constrained Component Analysis Metric Optimization Method Based on Difference Regularization

Abstract

The present invention relates to a pairwise constrained component analysis measurement optimization method based on difference regularization, comprising: step S1: using a camera to collect multiple pictures of pedestrians to form a training sample set, and extracting the color features of pedestrian targets in each sample picture to form a d-dimensional eigenvector x i , and finally form a training set containing N d-dimensional feature vectors, where N is the number of sample pictures; Step S2: Perform dimensionality reduction processing on the feature vectors in the training set; Step S3: Calculate the relationship between the reduced dimensionality samples Euclidean distance, and based on the Euclidean distance between each training sample and the test sample, the training samples are divided into positive samples and negative samples; Step S4: By adding a difference regularization term to the target optimization function, and creating an optimization based on constraints problem; step S5: use the gradient descent method to solve the optimization problem. Compared with the prior art, the invention has the advantages of avoiding the occurrence of over-fitting phenomenon, improving the generalization ability of the metric learning algorithm and the like.

Description

technical field [0001] The invention relates to the technical field of intelligent information processing, in particular to a pair-wise constrained component analysis metric optimization method based on difference regularization. Background technique [0002] In recent years, video surveillance systems have become widely popular. The construction and application of video surveillance systems are playing an increasingly important role in combating crime and maintaining stability. Video surveillance has become a new method for public security organs to investigate and solve crimes. In video reconnaissance applications, the retrieval of specific suspected targets is an important requirement. In the face of massive video data, relying entirely on manual methods to complete the above work not only consumes a lot of human resources, but also cannot guarantee the accuracy and real-time performance of the results. The operator's matching result is affected by individual experience...

AI Technical Summary

Problems solved by technology

[0005] To sum up, the distance measurement method is the focus of research in the field of pedestrian re-identification. Although many excellent research results have been achieved in domestic and foreign research, with the development of large-scale camera networks, the deterioration of shooting conditions, and the increasing security requirements Improvement, the performance of existing distance measurement algorithms in terms of pedestrian re-identification and matching accuracy cannot meet the needs of practical applications. Therefore, the research on distance measurement learning algorithms in this project will have very important theoretical significance and application value

Images