基于特征识别的群体运动轨迹分析算法  

An Algorithm for Analyzing Group Motion TrajectoryBased on Feature Recognition

作　　者：刘辉[1] LIU Hui(Physical Education Department,Modern College of Northwest University,Xi’an 710130,China)

机构地区：[1]西北大学现代学院,体育教学部,陕西西安710130

出　　处：《微型电脑应用》2024年第8期63-67,共5页Microcomputer Applications

基　　金：西安市2023年度社会科学规划基金项目(23TY103);西安市2019年度社会科学规划基金项目(19X08)。

摘　　要：针对群体运动轨迹分析算法存在的执行时间长、预测准确率低等问题,基于高斯混合模型、均值漂移算法和LSTM模型提出一种具有较高预测准确率的群体运动轨迹分析算法。通过回顾和分析主流的群体目标跟踪算法,设计和构建基于粒子滤波的均值漂移算法,并在此基础上引入具有时空预测功能的LSTM模型和高斯混合模型,从而实现视频环境下群体目标运动轨迹的抗扰识别、跟踪和预测。仿真实验采用相同的篮球运动视频数据集作为样本,其统计结果表明,与传统的群体识别和跟踪算法相比,基于特征识别的群体运动轨迹分析算法具有更高的预测准确率和更少的平均执行时间,群体运动轨迹的预测准确率提升了14.9个百分点,算法的平均执行时间降低了21.6个百分点。Aimed at the problems of long execution time and low prediction accuracy in group motion trajectory analysis algorithms,this paper proposes a group motion trajectory analysis algorithm with higher prediction accuracy based on Gaussian mixture model,mean shift algorithm and LSTM model.By reviewing and analyzing mainstream group target tracking algorithms,a mean shift algorithm based on particle filtering is designed and constructed.Based on this,an LSTM model and a Gaussian mixture model with spatiotemporal prediction function are introduced to achieve anti-interference recognition,tracking and prediction of group target motion trajectories in video environments.The simulation experiment uses the same basketball video dataset as an sample,and the statistical results show that compared with traditional group recognition and tracking algorithms,the group motion trajectory analysis algorithm based on feature recognition has higher prediction accuracy and less average execution time.The prediction accuracy of the group motion trajectory is improved by 14.9 percentage points,and the average execution time of the algorithm is reduced by 21.6 percentage points.

关 键 词：图像处理 目标检测 粒子滤波 均值漂移 运动轨迹 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TN21[自动化与计算机技术—计算机科学与技术]