空中视角下基于局部线性判别嵌入的目标跟踪方法  

Target tracking method based on local linear discriminantembedding in the air view

作　　者：陆艳[1] 孙锐[2,3] 王旭 Lu Yan;Sun Rui;Wang Xu(An Hui Broadcasting Movie and Television College,Hefei 230011,China;School of Computer and Information,Hefei University of Technology,Hefei 230601,China;Anhui Province Key Laboratory of Industry Safety and Emergency Technology,Hefei 230601,China)

机构地区：[1]安徽广播影视职业技术学院,合肥230011 [2]合肥工业大学计算机与信息学院,合肥230601 [3]工业安全与应急技术安徽省重点实验室,合肥230601

出　　处：《电子测量与仪器学报》2018年第11期90-96,共7页Journal of Electronic Measurement and Instrumentation

基　　金：国家自然科学基金(61471154);安徽省科技攻关科技强警项目(170d0802181)资助

摘　　要：空中视角下无人机的目标跟踪是在空中视角下辨认出运动目标,并且对目标进行准确且连续的追踪。如何实现在遮挡,尺度变化,实时性要求高的俯视条件下鲁棒的目标跟踪仍是亟待解决的问题。提出一种将采用局部线性判别嵌入(LLDE)特征提取的方法和核相关滤波训练分类器相结合的算法。LLDE是流形学习中的局部线性嵌入算法(LLE)的改进,通过在LLE的基础上构造了一个向量转换和距离标度模型,使得不同类的样本距离更远,同类样本距离更近,提高了分类精度有助于解决遮挡和尺度变化。同时核相关滤波训练分类器,能够快速的跟踪目标满足空中无人机跟踪的时实性要求。算法保持了相关滤波的快速跟踪算法实时性的优势,同时对遮挡和尺度变化有较好的稳健性。实验表明,与其他算法相比,算法在速度和精度方面有优势。UAV target tracking in air view is to identify moving targets from the perspective of the air, and keeps track of the target accurately and continuously. It is an urgent problem to quickly track the target and maintain the high robustness under occlusion condition, scale change and other complex environments. An algorithm adopting the local linear discriminant embedding(LLDE) feature extraction method and the kernel correlation filter training classifier was proposed. The LLDE algorithm is an improvement of local linear embedding algorithm(LLE) in flow learning. Based on LLE, the LLDE algorithm constructs a vector transformation and distance scale model, which makes the sample distance of different kinds of samples farther, and the similar sample distance closer, thus improves the classification accuracy and helps to solve the occlusion and scaling. At the same time, the kernel correlation filter training classifier can fastly track the target to meet the real-time requirement of air uav tracking. The proposed algorithm maintains the advantage of fast tracking algorithm of correlation filtering, and has good robustness to occlusion and scale change. The experiment shows that compared with other algorithms, the algorithm proposed in this paper has advantages in speed and accuracy.

关 键 词：流形学习 局部线性判别嵌入 核相关滤波 目标跟踪 

分 类 号：TN911.73[电子电信—通信与信息系统]