Whether and how is a surveillance camera jittering? A ROR perception based framework and method
Fei Gao,Kaitao Mei,Libo Weng,Yaozhong Zhuang
DOI: https://doi.org/10.1007/s10489-023-04631-3
IF: 5.3
2023-06-22
Applied Intelligence
Abstract:In the video surveillance, cameras often jitter slightly because of some uncertain factors like wind, loose of bracket, and so on, which may add more difficulties to subsequent intelligent video analysis. Then, it is great significant to detect whether and how is a camera jittering. In this paper, a concept, i.e. region of reference (ROR), is defined, and a framework and method of surveillance camera jitter detection is proposed through perceiving the location changes of the ROR. First, an extraction model of motion information is established to extract and eliminate the moving information such as moving objects, swaying leaves, etc. Second, global contrast and octree color quantization are employed to filter out the significant regions in the background, then, the most representative regions are chosen as the RORs by tracking the boundary of significant regions. Third, the RORs and their surrounding image features are extracted and input into the filter to construct the response map, then, the corresponding region of peak value in the response map is perceived for location change. Average accuracy of the proposed method on video jitter data set is 95.36%. The experimental results show that the proposed method can solve the problem of video jitter detection in complex scenes.
computer science, artificial intelligence