摘要
针对旋翼飞行机器人抖动造成运动图像模糊、行人跟踪时受部分遮挡等因素的影响,易出现行人目标丢失的问题,提出一种经过图像去模糊的图像帧序列的旋翼飞行机器人视觉跟踪方法。首先,为克服旋翼飞行机器人抖动造成运动图像模糊的问题,利用维纳滤波对运动图像帧序列进行去模糊处理;其次,为克服单一颜色描述子作为行人模型的缺陷,采取合适的权重融合行人的颜色和HOG描述子作为行人模型;考虑到Camshift算法能较好的跟踪无遮挡的行人目标,且实时性好,而粒子滤波(PF)算法对发生部分遮挡的行人目标的跟踪具有较强的鲁棒性,利用前后两帧图像的行人目标模型的相似度来实现Camshift算法和PF算法的切换,以保障算法实时准确的跟踪行人目标。利用维纳滤波对运动图像去模糊后,行人跟踪准确率从93%提升到100%;利用分段行人目标跟踪算法相比于Camshift算法更具鲁棒性,相比于PF算法,实时性提升26.45%,相比于跟踪学习检测(TLD)算法,实时性提升66%。行人发生部分遮挡时,旋翼飞行机器人跟踪行人的精度在50 cm以内,能够稳定跟踪行人目标。
The rotorcraft is vulnerable to attack by the blurred motion image caused by rotorcraft and partially occlusion of the pedestrian, resulting in the loss of the tracked target. To solve this problem, this paper proposes a segmentation pedestrian target tracking method for rotorcraft image deblurring. Firstly, in order to overcome the problem of moving images blur caused by rotorcraft shaking, Wiener filtering is used to deblur the moving image frame sequence. Then, in order to overcome the deficiency of the single color descriptor as a pedestrian model, the suitable weights are used to fuse the pedestrian’s color and HOG descriptors as pedestrian models. Considering the camshaft algorithm has good tracking effect and fast execution without occlusion, the particle filter algorithm can solve the tracking problem of partially occlusion with target, Camshift and particle filter algorithm are switched by using the similarity degree between the pedestrian target model of the two frames before and after, guaranteeing the robustness and real-time ability. Pedestrian tracking accuracy is increased from 93% to 100% after using Wiener filtering to deblur the moving image. The segmented pedestrian target tracking algorithm is more robust than the Camshift algorithm, Compared with the particle filter algorithm, the real-time performance is increased by 26.45%, and compared with the TLD algorithm, the real-time performance is increased by 66%.When pedestrian is partially obstructed,the rotorcraft tracking accuracy of pedestrian within 50 cm, and can steadily track pedestrian targets.
作者
王耀南
罗琼华
毛建旭
陈彦杰
周显恩
Wang Yaonan;Luo Qionghua;Mao Jianxu;Chen Yanjie;Zhou Xian'en(College of Electrical and Information Engineering,Hunan University,Changsha 410082,China;College of Mechanical Engineering and Automation,Fuzhou University,Fuzhou 350116,China;National Engineering Laboratory for Robot Vision Perception and Control Technology,Changsha 410082,China)
出处
《电子测量与仪器学报》
CSCD
北大核心
2018年第11期97-107,共11页
Journal of Electronic Measurement and Instrumentation
基金
国家自然科学基金项目(61573134,61433016,61471167)
国家科技支撑计划(2015BAF13B00)
湖南省科技计划(2017XK2102,2018GK2022,2018JJ3079)资助项目