Multi-view Data Aggregation for Behaviour Analysis in Video Surveillance Systems

摘要

Detecting restricted or security critical behaviour on roads is crucial for safety protection and fluent traffic flow. In the paper we propose an algorithm for the analysis of movement trajectory of vehicles using vision-based techniques. It works on video sequences captured by road cameras in multi-view mode. We integrate methods of background modelling, object tracking and homographic projection. Individual views are projected into a single, planar surface of road surface and then the detected movement path is compared with a template associated with an illegal movement. The effectiveness of the proposed solution is confirmed by experimental studies.