On the impurity of street-scene video footage

In this paper, we present the technical challenges facing researchers in developing computer vision techniques to process street-scene videos from the wild. Video footage captured by surveillance CCTV cameras and hand-held devices such as mobile phones and body-mounted cameras worn by police officers pose particular difficulties. Video formats are varied and often non-Standards compliant which leads to apparent corruption when rendered using standard players. Footage is low-quality either in resolution or in sharpness caused by free movement of the camera, fast panning and zooming or weather conditions. We describe our experiences working with the Metropolitan Police in London to find a solution to these problems and enable computer vision techniques to be used in the forensic analysis of videos in criminal investigations.