Improved OR-PCA for robust foreground detection
Yan Mo, Haixia Xu, Xiaoxue Ding · 2017
Efficient and accurate access to foreground objects in video is the basis and key in the field of motion vision. Its difficulty lies in the dynamic background, illumination changes and such complex situations. The foreground object detection method based on robust principal component analysis (RPCA) has made great progress. In recent years, the Online Robust PCA (OR-PCA) has effectively solved the shortcomings of complex calculation and poor real-time performance in traditional RPCA. However, there are still some defects, such as the low initial detection accuracy. In order to achieve better detection results, this paper presents an Improved OR-PCA via Batch Initialization method for robust foreground detection. The strategy of batch initialization in our work overcomes the problem of random initialization in OR-PCA, also avoids the defects of conventional batch processing. The comparison experiments prove the superiority of our method over the state-of-art methods in different cases.