Research On Human Joint Point Detection Algorithm Based On Deep Learning

Human joint point detection is a basic problem in computer vision,and has broad application prospects in the fields of behavior recognition,human-computer interaction,and pedestrian re-identification.Due to the rise of deep learning,research on human joint point detection algorithms has made great progress,but there are still some problems.The first problem is detecting the position of the human body,eliminating the background information in the natural scene,and reducing the difficulty in detecting the joint coordinates of the human joint in the scene.The second problem is how to establish a human joint point relationship model to improve detection accuracy.The third issue is considering the accuracy of the model and the amount of calculation of the model to improve the practicability of the model.This paper focuses on the above issues.Eliminating useless background information andreducing the difficulty of joint detection are important todetecting the position of the human body.Through the analysis of the existing algorithms,the YOLO algorithm is selected as the framework of the human detection algorithm.The deficiencies of the YOLO algorithm are improved by integrating the DenseNet network idea,adding the inception model and the spatiotemporal pyramid downsampling layer,and normalizing the target object frame of the loss function.Through comparison experiments,it is verified that the improved YOLO algorithm caneffectively remove most of the useless background information and improve the accuracy of human body position detection.Themodel of joint relationship between joints is based on human joint point detection.Aiming at the problem that the detection accuracy of the existing algorithm cannot be further improved because the different joint points are not considered,the human joint point detection algorithm based on mutual verification of joint points is proposed.The consciousness learning method and convolutional neural network are combined.The joint point detection is carried out in a gradual manner of easy and difficult.The multi-objective consciousness learning network model is used to carry out experiments on the LSP data set and the FLIC data set,and compared with other models.Experiments show that the models have achieved high detection accuracy on both data sets.Themodel established canimprove the detection accuracy.Reducingthe amount of model calculation is used toensure the accuracy and improve the practicability of the model.The existing single human joint point detection algorithms has a large parameters problem.Amulti-stage regression human joint point detection algorithm based on the scale-adjustable hourglass module is proposed in this article to solve the problem.The scale-adjustable hourglass module is used to construct the branch network at each stage to realize multi-scale feature learning.And the branch network can ensure the accuracy of the model to reduce the calculation amount of the model.The calculation amount of the model training phase and the test phase,the parameter quantity of the model and the accuracy of the model are given on the MPII and LSP datasets.It is verified that the proposed model can effectively reduce the calculation amount and ensure the accuracy of the model.