自适应双路间隔调控跟踪算法  

ADAPTIVE TWIN-CHANNEL INTERVAL REGULATION TRACKING ALGORITHM

作　　者：蒋秋萍 付小雁[1,2] Jiang Qiuping;Fu Xiaoyan(College of Information Engineering,Capital Normal University,Beijing 100048,China;Beijing Key Laboratory of Electronic System Reliability Technology,Beijing 100048,China)

机构地区：[1]首都师范大学信息工程学院,北京100048 [2]电子系统可靠性技术北京市重点实验室,北京100048

出　　处：《计算机应用与软件》2023年第1期173-183,共11页Computer Applications and Software

基　　金：国家自然科学基金项目(61876112,61603022)。

摘　　要：卷积特征可有效解决复杂场景下的跟踪漂移问题,但特征提取耗时较长;传统手工算法跟踪速度快,然而精度有待提高。基于此,提出自适应双路间隔调控算法。针对目标尺度变化问题,使用多尺度特征训练跟踪器并依据质量评估自适应融合结果,为减少计算开销采用主成分分析降维;算法实时评估目标运动状态,在手工主线算法基础上间隔性地调用深度算法,在保证跟踪精度的同时尽可能地减少时间消耗。实验表明该算法实现了特征优势互补,精度可达0.91,GPU速度17.5 FPS。Convolution features can effectively solve the tracking drift in complex scene, but it takes a long time to extract features. The manual algorithms have fast tracking speed, but the accuracy needs to be improved. Based on this, we propose the adaptive two-channel interval regulation tracking algorithm. Aimed at the scale variation of targets, the algorithm used multi-scale features to train the tracker and fused the results based on the quality assessment. In order to reduce the computational cost, principal component analysis was used to reduce the dimension. This paper evaluated the target state in real time, and applied the deep algorithm on the basis of the manual main line algorithm at intervals, which reduced the time consumption and ensured the tracking accuracy. The experimental results show that the algorithm achieves complementary advantages, and the accuracy of the algorithm can reach 0.91, and the GPU speed is 17.5 FPS.

关 键 词：相关滤波 多尺度 质量评估 深度卷积特征 双路间隔调控策略 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]