The Dutch National Police increasingly use mobile robots for their operations. Examples of applications are observation and surveillance. The robots that the police uses are commercial systems from a variety of manufacturers. All robots are equipped with a camera, which wirelessly streams video data to a tele-operator. The tele-operator uses the video to navigate the robot. In the future the tele-operator can be replaced by or assisted by an automated system that uses computer vision.
The wireless channel that is used to stream the video to the tele-operator will inevitably suffer from external disturbances as the robot moves around. Such disturbances lead to a lower throughput, which means that not all video data can be transmitted. The resulting videos therefore contain visible artefacts, making it difficult to navigate the robot. To solve this problem, the size of the video data can be reduced by discarding some of the data.
In this research three methods for reducing the video data are explored: spatial scaling, temporal scaling and quality scaling. Experiments are conducted where different combinations of these three types of scaling are applied to multiple videos. For each type of scaling, the reduction of required throughput is analysed. In addition, it is examined how well the resulting videos can be used by an automated system by analysing the performance of a visual tracking algorithm on the videos. The results can be used by manufacturers of mobile robots to optimize the configuration of the video stream based on available throughput.