Research On Action Recognition Method Based On Minimum Class Variance Of Approximate Kernel Extreme Learning Machine

Nowadays,human motion recognition has a long history,and under the premise of the rapid development of artificial intelligence,it has also received more and more attention,such as human-computer interaction,machine vision and content-based video retrieval.It has great applications in various fields,it can detect and recognize the movements of the human body,and finally recognize the action behavior of the target,and even predict the next behavior of the target,which is very important for the crime detection of the character.Since the detection target data set of the human motion recognition is mostly a camera with a large amount of data information,and even a data of a high-speed camera may be used,this brings a very large amount of calculation to the human motion recognition.In response to the appeal issue,we propose a new multi-view motion recognition method in this paper,which can solve the problem that the amount of video data is large and the recognition is difficult.First,this paper uses a fast and efficient motion feature extraction method called fuzzy feature extraction.Among them,what we have studied in this paper is the multi-angle human motion recognition method,and the fuzzy feature extraction has a non-good effect on the extraction of multi-view motion features.At the beginning,this method needs to blur the global picture of the video(ie,each picture in each action class in the video),and then perform k-means clustering on the fuzzified feature picture to generates our feature model for the training or recognition of the classification algorithm.In addition,we also propose a new algorithm for multi-view human motion recognition,which is improved by the extreme learning machine with very fast calculation rate.We call it the Minimum Class Variance of Approximate Kernel Extreme Learning Machine(MCVAKELM),The learning machine algorithm,which is an approximate core single hidden layer neural network obtained through training,aims to enhance the performance of motion recognition.Because the Extreme Learning Machine has many improvements,but in its development process,the slow computational complexity is increased,making it lose its original fast and efficient advantages.Therefore,the classification algorithm proposed in this paper can solve the problem of increasing computational complexity due to the large amount of appeal multi-view motion video data.Finally,we built an experimental model for performance evaluation.In this process,we used two commonly used human motion recognition data sets,such as KTH and UCF sports,to compare the performance of the improved support vector machine,traditional extreme learning machine and various improved extreme learning machines.And in the final test results,the proposed method has a better effect.