基于改进帧差法和Camshift算法的运动车辆检测与跟踪方法研究

发布时间：2019-01-29 06:07

【摘要】：智能视频监控越来越多地运用在社会各层人民的工作和生活中,其给人们带来的便捷不言而喻。在智能交通系统中,对运动车辆的检测和跟踪是整个交通事件检测和视频监控系统智能化的主要内容和关键技术。运动车辆检测的目的是将视频序列中感兴趣的区域无误地提取出来,这样才能为后续的跟踪提取相应目标的特征;运动车辆跟踪的目的是需要实时了解其运动特征,包括行驶方向、速度、轨迹等。做好车辆检测和跟踪的工作为交通事件检测、行为分析等奠定了坚实的基础。但是由于目标检测和跟踪的场景复杂,不可预测因素增多,目标因为摄像头的角度及距离等原因而大大增加了目标跟踪的难度。因此,开展运动目标检测及跟踪方法的研究具有重要意义。本文围绕高速公路运动车辆检测及跟踪方法开展了如下研究工作:(1)提出一种基于五帧差分和LOG算子相结合的运动车辆检测方法。首先在帧差类方法中,分别对二帧差分、三帧差分、四帧差分、五帧差分进行详细研究和仿真,通过实验结果对比得出五帧差分检测前景目标的效果最好,然后结合能够检测运动目标完整信息的LOG边缘检测算子进行运动目标检测。实验结果表明,两种算法结合的目标检测方法对环境的适应能力强,去噪效果良好,对目标检测信息完整,达到了方法改进的效果。(2)结合正则粒子滤波改善Camshift算法的运动车辆跟踪方法。在利用Camshift算法进行目标跟踪的过程中,其主要是根据所跟踪目标的色调分布来进行跟踪的,因此能实现对具有与周围环境相差较大的颜色的目标进行良好的跟踪。但是容易受到目标附近具有相似颜色的背景或环境干扰导致跟踪丢失,为此,本文融入正则粒子滤波方法提出了一种改善Camshift的运动目标跟踪方法,该方法通过与运动目标检测方法相结合,实现对新目标和相似目标的良好处理,实现跟踪的持续稳定,能较好地实现监控系统智能化。本文以高速公路运动车辆为研究对象,完成了基于帧差法和Camshift的运动车辆检测和跟踪方法的研究,丰富了监控视频中前景目标检测和跟踪的研究内容,对其他场景的前景目标检测和跟踪具有一定的借鉴意义。
[Abstract]:Intelligent video surveillance is more and more used in the work and life of people at all levels of society. In the intelligent transportation system, the detection and tracking of moving vehicles is the main content and key technology of the whole intelligent traffic incident detection and video surveillance system. The object of moving vehicle detection is to extract the region of interest from the video sequence unmistakably, so as to extract the features of the corresponding target for the subsequent tracking. The purpose of moving vehicle tracking is to know its motion characteristics in real time, including direction, speed, track and so on. The work of vehicle detection and tracking lays a solid foundation for traffic incident detection and behavior analysis. However, because the scene of target detection and tracking is complex and unpredictable factors increase, the target greatly increases the difficulty of target tracking because of the angle and distance of the camera. Therefore, it is of great significance to study the methods of moving target detection and tracking. The research work of this paper is as follows: (1) A moving vehicle detection method based on five-frame difference and LOG operator is proposed. First of all, in the frame difference class method, the two frame difference, three frame difference, four frame difference and five frame difference are studied and simulated in detail, and the results of experiment show that the effect of five-frame differential detection foreground target is the best. Then LOG edge detection operator which can detect the complete information of moving target is used to detect moving target. The experimental results show that the two algorithms have strong adaptability to the environment, good denoising effect, and complete target detection information. The improved method is achieved. (2) the moving vehicle tracking method based on Camshift algorithm is improved by using regular particle filter. In the process of target tracking using Camshift algorithm, it is mainly based on the hue distribution of the target being tracked, so it can achieve good tracking of the target with different colors from the surrounding environment. However, it is easy to lose tracking due to background or environmental interference with similar colors near the target. In this paper, a moving target tracking method to improve Camshift is proposed by incorporating regular particle filter. By combining with the moving target detection method, the new target and the similar target can be handled well, the tracking can be sustained and stable, and the monitoring system can be intelligentized. Based on frame difference method and Camshift, this paper studies the detection and tracking method of moving vehicle on expressway, which enriches the research content of foreground target detection and tracking in surveillance video. It can be used for reference for foreground target detection and tracking in other scenes.
【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U495;TP391.41

【参考文献】