Research On Pedestrian Re-identification In Based On Fusion Feature And Metric Learning

With the rapid development of computer vision and the popularization of i ntelligent monitoring equipment,pedestrian re-identification technology is more and more widely used.Pedestrian re-identification technology aims to match all images belonging to the same person in different real-time surveillance scenes,retrieve and identify pedestrian images in video surveillance,establish identity c orrelation information of target pedestrians,and provide technical support for in telligent security,intelligent search,intelligent business and automatic driving.H owever,there are many challenges in the real-time monitoring scene,such as e nvironment,occlusion and perspective change.On the basis of research at home and abroad,this paper starts from two main directions of feature expression and metric learning.From the above two aspects,a new feature fusion method and a measurement learning algorithm considering the nonlinear change under the cross view are proposed respectively.The main work of this paper is as follows:This paper expounds the background,significance,challenges and development trend of pedestrian re-identification technology.Research on Pedestrian re-identification based on multi-feature fusion algori thm.Retain the information of different features and combine different original f eatures.According to the size of Mahalanobis distance and Softmax function mo del to determine the weight fusion to get a number of new image features.The improved artificial fish swarm optimization algorithm is introduced to optimize the weight of the multiple similarity distances to obtain the similarity function with better performance.Research on Pedestrian re-identification based on metric learning algorithm.When processing high-dimensional small sample data,the estimation of the inverse of covariance matrix is usually prone to large deviation due to the small data set.Considering the complex nonlinear transformation of human appearance between different views,this paper introduces the minimum error classification and smoothing technology based on the nonlinear cross-view quadratic discriminant method,so as to realize the nonlinear measurement learning and effectively improve the estimation accuracy of the covariance inverse matrix.The validity of the feature extraction and measurement learning method proposed in this paper is verified.The algorithm performance is validated on multiple data sets.The experimental results show that both the feature extraction method and the metric learning method proposed in this paper can achieve better recognition accuracy and robustness.