SFD-SR: strengthened feature description and stabilized re-detection-based infrared small–dim target tracking algorithm
Zehao Li,Shouyi Liao,Meiping Wu,Tong Zhao
DOI: https://doi.org/10.1007/s00500-023-09307-1
IF: 3.732
2023-10-16
Soft Computing
Abstract:Infrared small and dim target has posed great challenges on the target detection and tracking, especially for early warning, navigation and guidance. However, most common tracking algorithms are inefficient for infrared small target due to huge computation, and complicated feature analysis on such target is useless and needless. As such, a fast and useful algorithm based on the kernel correlation filter (KCF) is proposed in this paper. It supplies a solution for target without clear characteristic, target missing and analog interference in complex backgrounds. The proposed algorithm focuses on model update and scale update rules, fusion feature expression and multi-cues constrained re-detection module; all of the ideas aim to optimize process and improve practicability. Firstly, the improved ridge regression model with solving boundary effect and model consistency can get more precise results. Secondly, a useful strengthened feature description (SFD) including the fused feature extraction and complete confidence indicator is proposed to improve the reliability of tracking. Thirdly, the stabilized re-detection (SR) strategy is designed based on motion rules, which can cope with abnormal disturbance and missing states. Fourthly, an adaptive model update is proposed, which can reduce the error accumulation. The experiment results and the comparative studies with some common algorithms show that the proposed algorithm achieves the success rate of tracking above 98%, RMSE of center location below 5 pixels and speed above 120 fps and tends to be more effective.
computer science, artificial intelligence, interdisciplinary applications