Research And Implementation Of Starting Action Correction System Based On Human Key Point Detection

As a modern sports power,my country,with the rapid economic development,the people’s material and cultural life quality has rapidly improved,and the concept of national fitness is deeply rooted in the hearts of the people.At the same time as the national fitness activities are launched,we have noticed that because of the high difficulty of some skill sports,people are prone to wrong actions when exercising,and even there are great potential safety hazards.In most cases,traditional sports training adopts the training method based on the naked eye observation of sports coaches.Sports coaches guide athletes by observing the movements of athletes during training and combining their own experience and opinions.There is no quantitative data for analysis.The training process is seriously affected by human subjective consciousness,which makes the training results of athletes vary greatly.In order to carry out scientific analysis and guidance on people’s movement posture,it is necessary to obtain the movement data or limb posture data of people when they exercise.In recent years,with the in-depth research on related technologies such as deep learning and human pose estimation,the application of these technologies in motion detection,motion data analysis,etc.enables us to obtain the limbs and related key point data of the exerciser,which not only can scientifically guide the entire training process of athletes,and at the same time improve the training efficiency of coaches and athletes,as well as the safety factor in sports.In this paper,a high-resolution human key point detection network based on attention mechanism is proposed,and an athlete’s starting motion correction system based on the key point detection network is designed and implemented.The three-part pose images in the running stage are taken as the main research object,and the application of computer vision technology in the key point detection of athletes’ starting pose images is further studied.The main contents of this paper are as follows:（1）The starting motion data set of athletes was established,and the starting images of the athletes in the three categories of the starting stage,the preparatory stage and the mid-running stage were collected through on-site shooting,network video interception and other methods,and the collected images were collected.The starting images are expanded through a variety of data enhancement methods and data standardization processing,and finally the images are classified and made into a data set of athletes’ starting sports for experimental use.（2）In this paper,through the study of traditional human pose estimation algorithms,it is found that most of the traditional pose estimation algorithms perform upsampling and downsampling in the network model,which leads to the loss of a lot of spatial information during training.In addition,the complex structure of the network model and the large number of parameters and computations also lead to low accuracy of the conventional network model for human pose estimation tasks.This paper studies the shortcomings and problems of traditional human pose estimation algorithms,and proposes a high-resolution human key point detection network based on attention mechanism.The introduction of the attention module is to make the network pay more attention to the target instance,and at the same time,try to control the amount of parameters and calculation to the maximum extent.This paper introduces three different attention modules into the network for ablation experiments,which proves the effectiveness of CBAM attention mechanism and model fusion for starting posture prediction.The experimental results show that the high-resolution human keypoint detection network based on the attention mechanism obtains a higher starting posture prediction accuracy,and effectively controls the amount of parameters and computation in the network training process.（3）An athlete’s starting motion correction system based on human body key point detection is designed and realized.The system mainly includes user registration and login function,posture image uploading function,image prediction function,display prediction results and correction suggestions function.