Human Activity Recognition Research Based On Metric Leraning

Human activity recognition is a popular research topic in the field of computer vision.Many applications,such as human-computer interaction,intelligent video surveillance,motion video analysis,patient monitoring systems,smart home and intelligent security systems,require accurate recognition of human activity categories in the video sequence.In order to accurately identify the categories of human activities,a suitable distance metric is required to represent the similarity between active samples.In particular,the traditional human activity recognition algorithm uses a custom distance metric to represent similarity.However,the actual application of different activities to identify tasks require different distance metric,so the use of manual way to construct the appropriate distance metric is extremely difficult.Considering the distance metric learning algorithm can according to different tasks,different data sets adaptively learning distance metric,this paper apply it to human activity recognition,in order to improve the precision of identification.The main contents of this paper are as follows:(1)Aiming at the problem of human activity recognition that the measurement of similarity between the samples,an activity recognition algorithm based on Fast Neighborhood Component Analysis is proposed.The algorithm learns a linear transformation in the input space by minimizing the classification error on the training set.Weizman data sets,Kernel1 data set and Kernel2 data set on the experimental results show that the distance metric Fast neighbor composition analysis algorithm learned can significantly improve the accuracy of human activity recognition.(2)Aiming at the problem of human activity recognition that the measurement of similarity between the samples,an activity recognition algorithm based on Large Margin Nearest Neighbor is proposed.The algorithm learns the Mahalanobis distance by formalizing the objective function into convex semidefinite programming problem.The experimental results show that the Large Margin Nearest Neighbor distance metric learning in common human activity recognition data sets obtained better recognition accuracy.