원문정보
초록
영어
Based on support vector machine (SVM) and analysis of optical flow, the paper presents a new method for recognizing player motions in broadcast sports video. The video often has problems like bad-quality image, non-static video cameras and low-resolution image of player. To address them, from the perspective of movement analysis and according to the spatial distribution features of optical flow field of tracked members, the grid classification method, a kind of local analysis idea, is used to extract descriptive characteristics of movement recognition. With different idea from the traditional flow analysis, the method regards optical flow vectors in the traced areas as a kind of spatial distribution information in the mode of mobility, improving robustness of optical flow features. With SVM as model classifier and the application of time-sequential voting strategy, the type of player actions is identified. Compared with existing recognition methods based on apparent characteristics, the proposed recognition technique which fetches and depends on motional descriptive feature achieves better recognition effects.
목차
1. Introduction
2. Player Tracking and Adjustment
3. Computation of Motion Descriptor
3.1. Optical Flow Estimation and Noise Erasure
3.2. Local Movement Analysis
4. Classification of Actions based on Supervised Learning
5. Experiment Design and Discussion
6. Conclusion
References