Video selection for visual sensor networks: A motion-based ranking algorithm

A Visual Sensor Network (VSN) is composed by several cameras, in general with different characteristics and orientations, which are used to cover a certain Area of Interest (AoI). To provide an optimal and autonomous exploitation of the VSN video streams, suitable algorithms are needed for selecting the cameras capable to guarantee the best video quality for the specific AoI in the scene. In this work, a novel content and context-aware camera ranking algorithm is proposed, with the goal to maximize the Quality of Experience (QoE) to the final user. The proposed algorithm takes into account the pose, camera resolution and frame rate, and the quantity of motion in the scene. Subjective tests are performed to compare the ranking of the algorithm with human ranking. Finally, the proposed ranking algorithm is compared with common objective video quality metrics and a previous ranking algorithm, confirming the validity of the approach.