基于视觉图像的田径失误动作识别仿真

Simulation of Recognition of Athletics Faults Based on Visual Image

对田径运动中的运动进行图像视觉准确分析,可以有效找到失误动作规律,指导训练。在高速的体育运动中,空间运动位姿测量过程需要高精度的图像特征点匹配,但是,在快速移动的图像运动特征中,特征匹配存在较大的滞后性,导致需要大量图像帧的差异特征与关键特征进行对比才能有效匹配,对后期动作的识别带来滞后性干扰。提出采用行为视觉算法,对田径失误技术对比分析,融合于局部保持映射的理论,保留田径失误技术图像初始特征的局部邻近结构组建失误技术初始特征向量,并对初始特征向量进行维数约简后,利用数字直脊变换理论对原始的技术失误图像进行重构,提取高频子带区域中可以体现失误技术细节变化的关键特征,有效的完成田径失误技术对比分析。仿真结果表明,改进方法可以提升运动图像识别效率。

Analyzing the visual image exactly can effectively find the rule of fault action to instruct training in the track and field sports. The measurement process of spatial motion posture needs to match with high - precision image feature points during fast - moving sports. In this paper, we proposed a theory which integrated with mapping pre- served partially and analyzed athletics fault by contrast. The initial feature vector of fault was established by partially reserving neighbourhood structure of fault image. After reducing the dimension of initial feature vector, we used the theory of digital straight ridge transformation to reconstitute the original fault image. Then we collected the key feature in the sub - band region with high frequency, which can reflect the detailed faults. The comparative analysis of athletics was effectively completed. The results of the simulation indicate that the modified method can improve the recognition ratio of motion images apparently.