视频监控中的运动车辆检测跟踪方法研究

发布时间：2018-06-24 22:48

本文选题：运动目标识别 + 光流法　；参考：《中国海洋大学》2014年硕士论文

【摘要】：近年来，随着与汽车、交通等相关技术的不断成熟和人民生活质量的大幅提高，城市化进程加快，我国家庭对汽车的拥有量急速增长，城市道路交通越来越拥堵，行车速度也越来越快，交通矛盾日益突出，交通事故频发，给人们的生命财产和国民经济造成了巨大损失。因此自动驾驶和安全驾驶越来越成为研究的热点，智能交通系统中车辆检测与跟踪技术的研究也越来越受到国内外学者们的高度关注。基于视频的运动物体识别算法一直是计算机视觉研究的热点。各国的研究人员对这个问题从不同角度给出了各异的解决方法，然而由于基于视频的运动目标识别的复杂性，在识别的实时性及识别精度上还有很多难点，还没有找到一种特定的方法适合所有的场景，特别是对于雨、雪、雾等非常规天气，对运动目标的识别仍然存在困难。本文首先分析了常用的基于视频的主要运动目标识别的研究意义与研究现状，并对常用检测算法进行了分析和比较，对特征提取和基于时域的跟踪方法进行了详细的介绍，对传统的检测跟踪算法同基于特征的光流法的优劣进行比较。接着对基于高斯分布的背景建模做了介绍，后面利用隐马尔可夫模型完成轨迹跟踪。最后，在结合国内外研究成果的基础上，，本文提出了一种适合于雾霾天的运动目标识别方法，该方法能在光线不好，遮挡严重的雾霾天较好的完成运动目标的识别。同时进行实验验证，并对实验结果进行分析。通过实验，可以验证基于特征的光流法及对运动目标进行低维特征提取的识别跟踪算法在实时性和准确性上有较好的表现。同时，也对目前仍存在的问题进行了分析，同时对下一步的研究方向提出展望。综上，本文提出的基于视频的运动目标识别方法，在确保识别精度的前提下提出了在恶劣的非常规天气下运动目标的识别方法，并验证了该方法的实用性。
[Abstract]:In recent years, with the continuous maturation of related technologies such as automobile, traffic and other related technologies and the substantial improvement of people's quality of life, the process of urbanization has been accelerated, the number of Chinese families owning cars has increased rapidly, and the urban road traffic has become more and more congested. The speed of driving is also getting faster and faster, the traffic contradiction is more and more prominent, the traffic accident frequently occurs, has caused the huge loss to the people's life and property and the national economy. Therefore, automatic driving and safe driving are becoming more and more research hotspot, and the research of vehicle detection and tracking technology in intelligent transportation system has been paid more and more attention by scholars at home and abroad. Video-based moving object recognition algorithm has been the focus of computer vision research. Researchers from different countries have given different solutions to this problem from different angles. However, because of the complexity of moving target recognition based on video, there are still many difficulties in real-time recognition and recognition accuracy. A specific method has not been found for all scenarios, especially for rain, snow, fog and other unconventional weather, so it is still difficult to identify moving targets. This paper first analyzes the research significance and research status of the main moving target recognition based on video, and then analyzes and compares the common detection algorithms, and introduces the feature extraction and time-domain tracking methods in detail. The advantages and disadvantages of the traditional detection and tracking algorithm are compared with the feature based optical flow method. Then the background modeling based on Gao Si distribution is introduced, and then the hidden Markov model is used to track the trajectory. Finally, based on the research results at home and abroad, this paper proposes a moving target recognition method suitable for smog days. This method can be used to recognize moving targets in poor light and severe smog days. At the same time, the experimental verification and analysis of the experimental results are carried out. The experiments show that the feature based optical flow method and the low dimensional feature extraction algorithm for moving targets have good performance in real time and accuracy. At the same time, the existing problems are analyzed, and the future research direction is prospected. In summary, the method of moving target recognition based on video is proposed in this paper, and the method of moving target recognition in bad unconventional weather is put forward, and the practicability of this method is verified.
【学位授予单位】：中国海洋大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：U495;TP391.41

【参考文献】