The Preliminary Study Of Basketball Video Analysis System

Nowdays, sports have become a frequently international competitive activity, which can partly show the economic strength of a country. In the last few years computer techniques have greatly improve the developments in many areas, but little has been done in sports.In this paper, computer graphics and digital image processing methods are implemented to analysis bascketball sports videos and generate several digital datas. According to these datas,. coaches can perform more accurate excercises for players and also can also get more reasonable clues in real-time game plays. In this paper, steps are provided to cover the whole stage from synchronous video capture to activity recognition.A new method is provided based on visual hull model, which can perform real-time object detection. First, front images are generated by background subtraction. Then, the visual hull algorithm is implemented to get the space information of each object in the scene and they can be separated fairly well.Because the players shade each other, common object tracking methods can hardly work well. A new tracking algorithm is built in this paper based on the object detection method above. To optimize the correct tracking process, space and color constrains are added here. The final algorithm is mainly derived from the probability model with a moving prediction. According to the experiment results, this algorithm can track multi-objects simultaneously with a high correct rate in realtime.Object detection method produces moving pathes for each player, but key problem is to recognize behaviors in bascketball games. Here, a method based on predefined templates. First, the court is divided into several areas and key elements are detected as the preprocess stage for game videos. Then, templates comparison can recognize several popular behaviors. Backerball experts have tested the prototype system based on this algorithm, and praised it for its sound accuracy and reference value.Finally, to prove the algorithms in this paper, two prototype systems are developed."GameMaster" system is for simultaneous multi-camera video capture and realtime analysis. The other system is a bascketball tactics animation editing software. By applying it, users can create bascketball animations as teaching materials easily and define behavior templates for activity recognition.Additionally, a B/S structured management information system is performed to collect and share basckerball materials as a platform on the network.