A rule-based evidence graph based video behavior verification method and system

Abstract

The application discloses a video behavior verification method and system based on a rule evidence graph. The method comprises the following steps: obtaining a video to be audited and a behavior rule package, analyzing the video to obtain a media index, and constructing a rule evidence graph; generating a first video processing plan according to the rule evidence graph, performing multi-modal recognition to generate video evidence events and constructing a multi-modal video evidence timeline; calculating rule proof scores and negative evidence coverage through graph matching, and determining the initial proof state of each rule node; generating a rule gap vector when the evidence is insufficient, increasing the sampling density in the candidate time window to generate a second video processing plan, updating the evidence timeline after execution, and recalculating the score, and outputting the final proof state. The application drives adaptive frame extraction and multi-modal joint verification through the rule evidence graph, combines the negative evidence coverage and the rule gap feedback mechanism, improves the short-time action recall rate and the verification accuracy, reduces the calculation cost, and realizes the traceability and reviewability of the verification conclusion.

Images