Design and Implementation of Scale Adaptive Kernel Correlation Filtering Algorithm Based on Hls
Xinyu Liu, Zhifeng Ma, Min Xie, Jiahe Zhang, Tingyan Feng · 2021
Moving target tracking is one of the important research subjects in the current field of computer vision, in which the correlation filter algorithms become a hotspot in theory research and application with its excellent comprehensive performance. However, the tracking frame of Kernel Correlation Filter (KCF) algorithms is fixed, and the change of target size will cause tracking drift or even target missing. This design improves it by integrating a scale filter from DSST (Discriminative Scale Space Tracking) algorithm so that it has a better Tracking effect under complex conditions such as target size change, occlusion, deformation and illumination change. At the same time, in order to balance the accuracy and real-time performance of the algorithm, the scale adaptive kernel correlation filtering algorithm is modified in this design, and an optimized design and implementation method based on high-level synthesis (HLS) is proposed to make it more suitable for FPGA.