复杂环境下的列车驾驶员目标检测

发布时间：2018-07-10 14:44

本文选题：复杂背景 + 列车驾驶员目标检测　；参考：《郑州大学》2017年硕士论文

【摘要】：从列车监控视频中自动、准确、快速地检测与定位列车驾驶员,已成为目前相关管理部门规范驾驶员操作行为,保证列车行驶安全的迫切需求。然而,在实际的监控视频中,由于图像分辨率较低、实际背景复杂、人体姿态多变、遮挡以及截断的情形较多,使得列车驾驶员目标检测成为了计算机视觉领域极具挑战性的研究课题。针对现有人体检测算法直接应用于列车驾驶员目标检测时所出现的问题,本文提出一种基于单张图像的列车驾驶员目标检测方法。首先,该方法以传统可变形部件模型为基础,分别提出单人检测器、遮挡检测器以及截断检测器,用以解决完整、遮挡、截断驾驶室场景下的人员检测难题;其次,综合三个检测器各自的特性,提出联合检测器,实现复杂行车环境下的列车驾驶员目标检测;最后,采用最优部件子集策略和coarse-to-fine策略分别从精度和速度两方面改进联合检测器,使其在提高检测精度的同时,也充分保证检测速度。联合检测器在基于单张图像的列车驾驶员目标检测中取得了较好的结果,但是该检测器并不能充分满足视频中的驾驶员目标检测需求,尤其是当驾驶员的肢体处于运动(包含微动)状态时,由于视频中的时空信息未被充分利用,在某些帧中往往会出现驾驶员误检及漏检问题。针对该问题,本文以联合检测器为基础,提出C-STC框架实现监控视频中的列车驾驶员目标检测。该框架首先利用联合检测器获取每帧图像的初始驾驶员目标检测结果;然后,使用空间上下文约束对每帧的初始检测结果进行后处理,抑制部分误检发生;最后,基于时间上下文约束策略,提出最优动态调整阈值方法实现视频中驾驶员的准确检测。实验结果表明,本文提出的联合检测器针对单张图像实现了准确、快速的列车驾驶员目标检测。在此基础上,综合联合检测器与时空约束策略的共同作用,使得本文所提出的C-STC框架针对监控视频的列车驾驶员目标检测取得了较好的检测结果,并可应用于实时系统中。
[Abstract]:Detecting and locating train drivers automatically, accurately and quickly from the video of train surveillance has become an urgent need for the relevant management departments to regulate the driver's operation behavior and ensure the safety of the train running. However, in the actual surveillance video, due to the low resolution of the image, the actual background is complex, the human body posture is changeable, occlusion and truncation are more, Train driver target detection has become a very challenging research topic in the field of computer vision. In order to solve the problem that the existing human body detection algorithm is directly applied to train driver target detection, this paper presents a method of train driver target detection based on single image. Firstly, based on the traditional deformable component model, the single detector, occlusion detector and truncation detector are proposed to solve the problem of personnel detection in the scene of complete, occluded and truncated cab. Based on the characteristics of the three detectors, a joint detector is proposed to detect the target of train driver in complex driving environment. Finally, the optimal component subset strategy and coarse-to-fine strategy are used to improve the accuracy and speed of the joint detector, respectively. So that it can improve the accuracy of the detection, but also fully ensure the detection speed. The joint detector has achieved good results in train driver target detection based on single image, but this detector can not fully meet the needs of driver target detection in video. Especially when the driver's limbs are in the state of motion (including fretting), because the space-time information in the video is not fully utilized, the problem of the driver's false detection and missed detection often occurs in some frames. In order to solve this problem, a C-STC framework is proposed to detect train driver targets in surveillance video based on joint detector. The framework firstly uses a joint detector to obtain the initial driver target detection results of each frame image; then, the spatial context constraints are used to post-process the initial detection results of each frame to suppress the occurrence of partial misinformation. Based on the temporal context constraint strategy, an optimal dynamic threshold adjustment method is proposed to detect drivers accurately in video. The experimental results show that the proposed joint detector realizes accurate and fast train driver target detection for a single image. On this basis, the combined action of joint detector and space-time constraint strategy makes the proposed C-STC framework obtain better detection results for train driver target detection in surveillance video, and it can be applied to real-time system.
【学位授予单位】：郑州大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U29-39;TP391.41

【相似文献】