Research On Human Object Behavior Recognition Algorithm Based On Combine Deep Learning With Principal Component Analysis Network

As an important branch of computer vision and pattern recognition,behavior recognition of human object is one of the trending topics in current research,which is related to daily life,life and property safety,social harmony and stability.At present,the problem of human object behavior recognition in simple scenes has been basically solved.However,how to quickly and accurately recognize the object in a complex video-based environment has become the focus and hot topic of current researchers.Based on the research results in the field of computer vision at home and abroad,On the basis of the analysis and research of existing human object behavior recognition methods,this paper presents human object behavior recognition algorithm based on deep learning aided principal component analysis network.The main idea of the algorithm is to use deep learning instead of the traditional foreground extraction method to detect human objects in complex environment and utilize principal component analysis networks to realize the human object behavior recognition.The research work of this paper is as follows(1)Overall algorithm design.Based on the analysis of existing human object behavior recognition method,a framework of human object behavior recognition algorithm based on deep learning aided principal component analysis network is presented.The framework consists of two parts: human object detection and human behavior recognition in complex scenes.Among them,the first part is to use the deep learning algorithm model to detect and extract the human object of the video image,and remove the unrelated noise data in the complex background and image.The second part is to put the extracted human object data with specific behavior into the principal component analysis network model trained by the behavior dataset for behavior recognition.(2)Research on human object detection methods in video based on deep learning.For the problem of object foreground extraction and detection,this paper uses the deep learning object detection model to extract and detect the human objects in complex scenes.In addition,in order to improve the object detection rate of the model,the object detection model is optimized using the focus loss principle and the dense network theory.The experimental results show that the optimized model can obtain more accurate feature information and improve the detection rate of human object in complex scenes while ensuring the object detection speed.(3)Research on human object behavior recognition method based on principal component analysis network.Aiming at the problems of high computational complexity and low computational efficiency of human object behavior recognition methods at present,combining with the theory of 3D convolutional neural network,an unsupervised human object behavior recognition algorithm based on 3D convolutional principal component analysis network is presented.In the network,the main component feature is used as a "convolution kernel" to filter the image,thereby image depth feature information can be acquired.The experimental results show that the algorithm has a good performance in improving training efficiency and recognition accuracy.(4)Algorithm implementation and analysis.This paper uses the open source about video image behavior dataset to train and test the behavior recognition algorithm based on deep learning aided principal component analysis network.The test results show that the recognition accuracy of the three types of dataset in KTH,YouTube_Action and UCF_sports which are 94.5%,86.7% and 92.4% respectively,as well as verify that the algorithm has high availability in behavioral video.