改进DTW算法的游泳训练视频手臂动作识别方法

A Method for Recognizing Arm Movements in Swimming Training Videos Based on Improved DTW Algorithm

针对复杂场景下游泳训练视频不同手臂动作识别准确性不足的问题,研究一种基于改进动态时间规整(DTW)算法的游泳训练视频手臂动作识别方法.将图像划分为若干个超像素区域,并通过计算显著值将像素合并成显著区域;借助人体骨骼结构,从显著区域中提取游泳手臂动作特征参数;将特征组成时间序列数据,利用改进DTW算法通过相似度匹配的特性来完成动作类型识别.结果表明:该方法的杰卡德系数较高,可以识别和分析不同场景下不同泳姿动作,具有更高的准确性和一致性.

In response to the problem of insufficient accuracy in recognizing different arm movements in swimming training videos in complex scenes,this article studied a swimming training video arm movement recognition method based on an improved DTW algorithm.In complex scenarios,the recognition ability for different swimming postures or movements was limited.To address the above issues,a swimming training video arm motion recognition method based on an improved DTW algorithm was studied.The image into several superpixel regions and merge the pixels intosalient regions by calculating salient values,extracting swimming arm motion feature parameters from salient regions using human skeletal structure.Features were combined into time series data and the improved DTW algorithm was used to achieve action type recognition through similarity matching.The results indicated that the studied method had a high Jaccard coefficient,which can identify and analyze different swimming postures in different scenarios,with higher accuracy and consistency.