基于融合特征多尺度的抗遮挡核相关滤波算法

Multi-scale anti-occlusion kernel correlation filtering algorithm based on fusion features

针对传统相关滤波算法进行改进,以提高算法在目标发生尺度变化、遮挡形变等复杂场景时的跟踪性能,提出一种融合尺度自适应和重检测机制的鲁棒性能的跟踪算法。该算法在融合FHOG和CN两种互补特征基础上,引入一种尺度自适应策略解决了尺度变化的问题,此外还进一步优化了模型更新策略并加入重检测机制,增强算法鲁棒性能。通过OTB100数据集测试结果表明,所提出算法相对于KCF算法精确度和成功率分别提升4.9%和17%,平均跟踪速度为45帧/s,且在遮挡、尺度变化和光照变化等场景下表现优异,能有效实现长期跟踪目标。

The traditional correlation filtering algorithm is improved to improve the tracking performance of the algorithm in complex scenes such as scale change and occlusion deformation.A robust tracking algorithm combining scale adaptation and re detection mechanism is proposed.Based on the fusion of FHOG and CN complementary features,a scale adaptive strategy is introduced to solve the problem of scale change.In addition,the model updating strategy is further optimized and the re detection mechanism is added to enhance the robustness of the algorithm.The OTB100 dataset test results show that the accuracy and success rate of the proposed algorithm are improved by 4.9%and 17%,respectively,compared with KCF algorithm.The average tracking speed is 45 frames/s,and the performance is excellent in occlusion,scale change and illumination change scenes,which can effectively achieve long-term target tracking.