Eigen Value Difference Based Technique for Shot Boundary Detection in Sports Videos

The goal of video segmentation is to partition a given video sequence into smaller meaningful units, based on temporal changes in the video sequence. The task of shot boundary detection is fundamental step of video segmentation and detects the boundaries of meaningful units of video sequence called shots. This is achieved by detecting temporal discontinuities in visual content in terms of some extractable features of frames. The main issues in video segmentation are the choice of features that are robust to illumination and camera/object motion and measure of dissimilarity or distance measure for detecting temporal discontinuities. Another complication is determination of threshold value for distance measure. This work proposes a novel technique using eigen value difference of video frames as a feature for measuring discontinuity to determine shot boundaries in sports videos. Further an adaptive thresholding methodology is also proposed to detect the boundaries. The test results shows that performance of the technique is superior to traditional techniques in terms of accepted performance parameters such as recall and precision.