Double Peaks Detection for Object Tracking

The kernel correlation filter tracker (KCF) predicts the object position based on the maximum value of the response map, but the response value of the real target is not always the maximum value of the response map. An improved kernel correlation filter tracker based on double peaks detection is proposed. Firstly, the peak-to-sidelobe ratio (PSR) of the target was calculated according to the response map obtained by the KCF algorithm. When the PSR is smaller than the empirical threshold, the highest peak and the second highest peak were detected in the response map, and then the PSR of response maps were calculated respectively in peaks. The position of the target was determined by bigger PSR. Finally, the template adopted an adaptive update strategy and the update process of the target template was optimized by PSR of response map. Experimental results show that the improved KCF increased the distance accuracy by 6.9% and the success rate by 5.5% with less time consumption.