Footwork Recognition And Trajectory Tracking Of Basketball Players Based On Multi-sensor

With the increasing abundance of current material life,people begin to pursue a healthier lifestyle.Sports represented by basketball are effective ways to improve physical function.With the increasing popularity of basketball,more and more scholars focus on the use of digital and systematic management methods to further improve the safety and effectiveness of basketball.The development trend of digital sports in the future is to establish a complete and reliable basketball data analysis system to analyze and guide the players.However,most of these systems are based on video monitoring technology at present.Their actual operation processes are limited by the deployment environment and they have disadvantages such as high cost.In order to achieve the widespread popularity of basketball analysis,this thesis utilizes a multi-sensor system based on intelligent wearable devices to collect the basic data of players.And information mining is carried out to realize footwork recognition and track tracking of basketball players through advanced data processing algorithms such as artificial intelligence.The main work and contributions of this thesis are summarized as follows:(1)This thesis collects preliminary data information such as the acceleration and angular velocity of basketball players’ footwork movements by means of several sensor devices embedded in smart insoles.Five kinds of footwork of basketball players,i.e.,sideslip step,back step,cross step,jab step and jump step,are recognized and classified combined with the structure of convolutional neural network(CNN).The performance of the neural network is optimized by adjusting super parameters such as the size of convolution kernel and the convolution step.The designed algorithm model obtains higher recognition accuracy compared with KNN and SVM.(2)To improve the computational efficiency and accuracy of the convolutional neural network in identifying footwork of basketball players,the dual-model convolutional neural network(DMCNN)structure is designed.The high-dimensional data collected by sensors are segmented according to the data features and used as the input of the two convolutional neural networks to form a reasonable mapping relationship with the output features.In addition,in order to optimize the training time of DMCNN,the principal component analysis(PCA)method is used to reduce dimension of feature vectors.The experimental results show that the performance of DMCNN is better than that of a single CNN structure.(3)In order to obtain the motion trajectory of basketball players,this thesis uses the sensor array in smart insoles to collect raw data.Combined with quaternion and the complementary filtering algorithm,the rotation matrix is established which realizes the data conversion from the sensor coordinate system to the reference coordinate system.Finally,the motion trajectory can be effectively restored through drift correction of the data.Experimental results show that this scheme has a small displacement error when realizing the trajectory tracking of the basketball player.