Using Feature Tree Model to Track High Speed Flying Soccer in Complicated Background

In this article, We try to analyze the video of soccer shooting, Where contains the key messages: position, speed and the flying curve. The curve is the most difficult message to get. K-means clustering is the common algorithm. But it needs a mass storage space and calculating time Which affects the real time performance. In the article, a feature tree model based on K-means clustering for moving object detection and recognition is put forward. The algorithm need not to build different feature clustering for different templates of soccer. (1) An union HOG feature and FLOW feature extraction method is put forward, including the character representation. (2) A K-means clustering feature construction method is put forward, Which can reduce the feature times effectively and improve the real time performance by means of feature tree construction. Moreover, after the recognition of the soccer in space coordinates, We generate the final curve in neighborhood comparison method.