Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
407852 | Neurocomputing | 2014 | 13 Pages |
•Moving object detection is the initial step of information extraction for video applications.•An encoder combined moving object detection method is proposed.•The proposed scheme achieves superior processing speed with competitive detection accuracy.
Moving object detection is the initial step of information extraction for video surveillance applications. For efficient transmission and storage, surveillance videos are compressed after capture. Existing methods usually take the compressed videos for moving object detection, which need full or partial decoding. Thus, they are not very efficient. Actually, the data generated during video encoding can be shared with detection, which avoids extra decoding. In this paper, we propose a low computation moving object detection method combined with video encoder. To avoid extra decoding, data generated during video encoding, such as Rate Distortion Cost (RDCost), are utilized for moving object detection. Experiments on a number of video sequences demonstrate its effectiveness in detection accuracy and superior processing speed.