T. Lertrusdachakul, T. Aoki, and H. Yasuda (Japan)
Camera motion, video retrieval, feature detection, feature tracking, speed analysis
In this paper, we evaluate the improvement in efficiency of video retrieval system due to the camera work information. The camera information is analyzed based on the pattern analysis of motion trajectories of image features. The algorithm generates the metadata of camera motion (i.e., panning, tilting, zooming, static, and the combination of panning and tilting) and its speed for the video sequences. The speed of camera motion is translated into the degree of human perception level (i.e., fast, slow, and medium). By applying the proposed analysis, we implement the camera work based video retrieval system. With this system, the users are able to search for a target shot by input the queries of camera motion and its speed. We compare the time used to search for the target shot of the normal mode and fast mode with the addition of camera information. The experimental results show that the camera work based shot retrieval system is significantly faster than the normal video search.
Important Links:
Go Back
