Frame with significant perceived motion energy for mobile visual sensing system

Visual sensors are usually set up to capture the environmental information and/or to monitor traffic road condition. The captured video is then transmitted to end-users for being further analyzed or stored. However, the video data amount and size is too large to store and transmit smoothly with high resolution under capacity and bandwidth limitation. Because consecutive frames have a large-scale overlapped range in content, not all frames are needed to be compressed and transmitted. In order to solve the problem of transmitting too much redundant video content to end-users, an effective video data reduction based on the perceived motion energy (PME) value is proposed to select the key-frames to represent and summarize the whole video. Five different videos consisting of standard tested, indoor, and outdoor environments are used to demonstrate the outstanding performance of key-frame extraction algorithm. For those videos with simple content and identical background, the reduction ratio is more than 60%. Moreover, after the video data reduction process, the system performance can be still maintained. In the end, one set of videos for visual odometry used in mobile robots is also tested. The overall reduction ratio of using proposed PME-based video data reduction can be more than 30%.