Efficient annotation of traffic video data

This paper presents a software application to generate ground-truth data on video files from traffic surveillance cameras used for Intelligent Transportation Systems (IT systems). The computer vision system to be evaluated measures the number of vehicles that cross a line per time unit --intensity-, the speed and the occupancy. A typical scenario is a camera on a pole 5 to 12m high pointing to the street in a city or to the lanes in a motorway. The application presented here is a tool to navigate through the video and annotate each instant when a vehicle crosses the target line and other features like its speed. The main target of the visual interface presented in this paper is to be easy to use, and with easy to find and non-specific hardware. It is based on a standard laptop or desktop computer and a Jog shuttle wheel, affordable and very common in Broadcast Video Edition. The setup is efficient and comfortable because one hand of the annotating person is almost all the time on the space key of the keyboard while the other hand is on the jog shuttle wheel. The mean time required to annotate a video file ranges from 1 to 5 times its duration (per lane) depending on the content.