In order to improve the accuracy of long jump in long jump, combined with computer vision image processing method to correct the long jump trajectory in long jump, an adaptive tracking method of long jump trajectory tracking image based on machine vision tracking detection is proposed, and the video point frame scanning method is used to collect the long jump trajectory tracking image. The image of long jump athletes is segmented by adaptive pixel fusion method, and the automatic tracking and recognition of long jumpers’ motion trajectory tracking image is carried out based on dynamic feature segmentation. The grey feature quantity of long jump trajectory tracking image is extracted, and the neighborhood distribution model of long jump in long jump is constructed. According to the dynamic evolution characteristic distribution of the long jump trajectory, the dynamic characteristics of the long jump trajectory are analyzed, and the image segmentation of the long jump track tracking is realized by combining the spatial neighborhood enhancement technology, and the adaptive tracking of the long jump trajectory in the long jump is realized according to the image segmentation results. The simulation results show that this method has high accuracy in adaptive tracking image of long jump athletes, and improves the accuracy of long jump in long jump.

In order to improve that segmentation quality of the video image of the marathon, a video image segmentation algorithm based on machine learning is proposed. Constructing the edge contour feature detection and the pixel feature point fusion reconstruction model of the marathon moving video image, carrying out multi-level feature decomposition and gray pixel feature separation of the marathon moving video image, and establishing a visual feature reconstruction model of the marathon moving video image, the feature segmentation and the edge contour feature detection of the marathon moving video image are carried out in combination with the block area template matching method, the similarity information fusion model is used for carrying out the video information fusion awareness and the block area template matching in the process of the marathon moving video image segmentation, the fuzzy feature quantity of the moving video image of the marathon is extracted, and the machine learning method is adopted to realize the fusion awareness and the segmentation quality evaluation of the marathon moving video information. The simulation results show that the method is good in image segmentation quality and high in image recognition, and the output signal-to-noise ratio of the motion feature reconstruction of the marathons moving video is high.