复杂场景下基于改进DiMP算法的精确目标跟踪  

Accurate object tracking based on improved DiMP algorithm in complex scenes

作　　者：忻瑶 韩华[1] 王春媛 熊雨滋 许莹莹 XIN Yao;HAN Hua;WANG Chunyuan;XIONG Yuzi;XU Yingying(School of Electronic and Electrical Engineering,Shanghai University of Engineering Science,Shanghai 201620,China)

机构地区：[1]上海工程技术大学电子电气工程学院,上海201620

出　　处：《智能计算机与应用》2022年第12期82-87,92,共7页Intelligent Computer and Applications

基　　金：国家自然科学基金(61305014);上海市自然科学基金(22ZR1426200)。

摘　　要：针对DiMP目标跟踪算法在自然场景下遇到遮挡及背景干扰导致跟踪表现不佳的问题,提出了改进的DiMP精确目标跟踪算法。在图像预处理阶段创新性地设计了一个任意灰度块替换策略来丰富样本的信息;将特征提取网络ResNet-50提取的目标各阶段的特征图输入到设计的多尺度融合模块中进行正向和反向的充分融合,得到包含更多位置信息和语义信息的特征图;随后特征图输入到模板预测模块中进行在线更新操作,进而得到判别力更强的目标模板。实验表明:该算法在UAV123数据集的遮挡和背景干扰测试中的成功率和精确率分别提高8%、4.15%和9%、6.30%;同时,在VOT2018的EAO指标上提高1.36%,在UAV123的成功率和精确率指标上分别提高3.89%和3.06%。说明改进的DiMP算法在对遮挡与背景干扰问题上优势明显,进而提升了算法的整体表现。Aiming at the problem that DiMP object tracking algorithm encounters occlusion and background interference in natural scenes, which leads to poor tracking performance, an improved DiMP accurate object tracking algorithm is proposed. In image preprocessing stage, an arbitrary gray-scale block replacement strategy is innovatively designed to enrich the information of samples;the feature maps of each stage of objects extracted by ResNet-50 are input into designed multi-scale fusion module for full fusion of forward and reverse directions, and feature maps containing more location information and semantic information are obtained;then feature map is input into template prediction module for online update operation, and target template with stronger discriminative power is obtained. Experiments show that success rate and precision rate in the occlusion and background interference tests of UAV123 dataset are increased by 8%, 4.15% and 9%, 6.30%, respectively;meanwhile, EAO indicator of VOT2018 is increased by 1.36%, success rate and precision rate of UAV123 are increased by 3.89% and 3.06%, respectively. The simulation shows that improved DiMP has a higher advantage in occlusion and background interference problems, thereby improving the overall performance of the algorithm.

关 键 词：替换策略 多尺度融合模块 DiMP 目标跟踪 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]