Detection of Moving Object with Dynamic Mode Decomposition and Yolov5
Zijian Chen, Jihua Lu, Xu Liu, Yan Lei · 2022 41st Chinese Control Conference (CCC) · 2022
Object detection is essential to timely restore the video, especially over the complex underwater environment. We propose a novel moving object detection algorithm with dynamic mode decomposition (DMD) and Yolov5. First, DMD is exploited to compress and retrieve the dynamic foreground and the static background by decomposing the snapshot sequence matrix into a low-rank and a sparse matrices, respectively. Then, the foreground video is retrieved from the low-rank matrix and the reconstructed matrix is recovered together by the two matrices. Finally, the moving object buried in the dynamic foreground and reconstructed images or videos are recognized by Yolov5. Experiments reveal that both the foreground and the reconstructed videos have higher detection accuracy than the Yolov5. Also, compared with the original video, the proposed algorithm bears the advantages of improved detection accuracy, lower compressing rate and decreased computing cost.