A Cost-Effective and Robust Shot Change Detection and Classification Technique based on Camera Tracking

Shot is considered a good unit for indexing and retrieval in video databases. Existing automatic shot boundary detection techniques require many threshold parameters which are difficult to select. To address this drawback, we present in this paper a content-based approach based on camera tracking. This is accomplished by tracking the background areas of the video. This strategy can detect and identify both abrupt and gradual changes. Our experimental results, based on six categories of 26 videos, indicate that the proposed method is more reliable than existing schemes although it processes 400% less data.