GPU-accelerated abrupt shot boundary detection
Youxian Zheng, Yuan Zhang · 2016
The detection of abrupt shot boundary is a fundamental task of video analytics and content-based video retrieval. The traditional methods tend to take much time in frame processing. In this paper, a GPU-accelerated abrupt shot boundary detection algorithm is proposed. This algorithm takes into account of both global feature and local feature of the video frames, in which the block HSV histograms and SURF are accelerated on GPU. We also adopt the GPU-accelerated Support Vector Machine (SVM) to identify the abrupt transitions. With the combination of global and local features, the algorithm is more robust against the abrupt camera/object motion and flash lights. Meanwhile, with the GPU acceleration, the proposed method achieves high speed with high accuracy as well.