摘要
为解决复杂环境下跟踪算法面临的光照变化、遮挡及相似干扰等问题,提出了基于多模型融合的自适应分块鲁棒目标跟踪算法,算法基于局部敏感直方图构建光照不敏感特征,基于超像素分割对采集图像进行自适应分块,以充分发挥子块的特征一致优势;通过相对熵和均值聚类构建双权值约束,以自适应提取高置信度子块进行目标跟踪,从而避免全局搜索并提高跟踪的精确性和实时性;算法通过遮挡检测和子块异步更新,进一步提高算法的运行效率和鲁棒性,提高模型对背景信息的过滤能力。实验结果表明,与其他跟踪方法相比,在包括遮挡、光照变化、相似目标和背景干扰等复杂场景下,该算法具有更优的跟踪精度和跟踪正确率,以及对不同场景的适应性和鲁棒性。
To solve such problems as illumination changes,occlusion and similar interference faced by tracking algorithms in complex environments,an adaptive moduler robust target tracking algorithm based on multi-model fusion is proposed. The light-insensitive features are constructed based on the local sensitivity histogram,and the acquired images are made adaptively partioning based on super-pixel segmentation to give full play to the consistent advantages of sub-modules. And then,dual weight constraints are constructed through relative entropy and mean clustering,and the high-confidence submodules are adaptively extracted for target tracking,so as to avoid the global search and to improve the accuracy and real-time of tracking. Finally,the tunning efficiency and robustness of the algorithm are further improved through occlusion detection and asynchronous update of sub-modules,and the ability of the model to filter background information is improved. The experimental results show that,compared with other tracking methods,the proposed algorithm has better tracking precision and correction rate,in such complex scenes as occlusion,lighting changes,similar targets and background interference,etc.The algorithm has adaptability and robustness to different scenes.
作者
欧阳美龙
OUYANG Meilong(College of Engineering and Technology,Eastern Liaoning University,Dandong 118000,China)
出处
《火力与指挥控制》
CSCD
北大核心
2022年第9期120-127,136,共9页
Fire Control & Command Control
关键词
目标跟踪
光照不敏感特征
超像素自适应分块
遮挡检测
异步更新
target tracking
light-insensitive features
super pixel adaptive partioning
occlusion detection
asynchronous update
