Abstract:
The boundary effect problem can be caused when the correlation filtering tracking algorithm uses the cyclic matrix to generate training samples for correlation detection. The farther the target is deviates the center of the detection area, the more serious the boundary effect problem will be. In order to solve the above problems, the method of correcting correlation region for background-aware correlation filter tracking based on BACF algorithm is proposed. The method uses different multi-feature fusion methods for feature extraction for grayscale and color videos. For new video frames, edge contour detection algorithms are used to correct the range and center position of the correlation detection area, so that the correlation detection area of the tracked target more reasonable, let the tracked target be located in the center of the correlation detection area as much as possible, so as to achieve the effect of alleviating the boundary effect. The experiment uses OTB100 data sets. The experimental results show that the improved algorithm effectively promotes the accuracy and success rate performance of the basic tracking algorithm.