Keyframe labeling technique for surveillance event classification

摘要: The huge amount of video data generated by surveillance systems necessitates the use automatic tools for their efficient analysis, indexing, and retrieval. Automated access to semantic content videos detect anomalous events is among basic tasks; however, due high variability audio-visual features large size input, it still remains a challenging task, though considerable research dealing with automated has appeared in literature. We propose keyframe labeling technique, especially indoor environments, which assigns labels keyframes extracted detection algorithm, hence transforms input an event-sequence representation. This representation used unusual behaviors, such as crossover, deposit, pickup, help three separate mechanisms based on finite state automata. are detected grid-based motion moving regions, called appearance mask. It been shown through performance experiments that algorithm significantly reduces storage requirements yields reasonable event classification performance.