Automatic detection of abnormal vehicle behavior within large-scale aerial video data
Reconnaissance and surveillance tasks such as scene understanding and detection of abnormal behavior are difficult to accomplish, as usually large amounts of video data need to be analyzed. Aerial vehicles with long flight duration and new sensors that are able to acquire high-resolution wide area videos increase the amount of aerial video data even more and generate the demand for efficient data handling approaches. In this paper, we focus on the detection of abnormal behavior of moving vehicles inside a region of interest that is chosen by a human operator and given by GPS coordinates. Abnormal behavior is for example a fast driving vehicle or a vehicle driving in the wrong direction. The proposed processing chain incorporates modules for efficient video database handling, moving vehicle detection, track fusion with different data sources and, finally, the detection of abnormal behavior. All presented modules enable interoperability by using STANAG-conform interfaces, i.e. STANAG 4607, 4609 and 4676.