事故车辆图像特征点的自动匹配

实现交通事故现场勘查中物体的立体视觉量测，需要对勘查获取的立体图像进行匹配。但由于一般自动匹配记录信息和事故研究所需信息的差异，使得自动匹配成果难以直接应用到交通事故勘查的立体匹配中。针对该问题，研究提出了一种事故车辆图像特征点的自动匹配方法，应用SIFT理论和BP神经网络理论，构造了事故车辆图像特征点的自动匹配模型，并进行了实验验证。研究结果表明，该匹配方法可以实现立体像对中任一选定目标的自动匹配，匹配的结果能够做为精匹配的提示。该项研究大幅度提高立体匹配选点的效率，加快交通事故立体视觉系统量测的速度，还可以作为一种技术手段给车牌被遮挡的违章车辆的确认提供借鉴。

Auto Matching Image of Crashing Vehicle

In order to go on stereo vision measure on traffic accident scene investigation, it is necessary to match image. But the common auto matching is difficult to meet for the special demand, because of the difference between the points the study concerning on and the points the computer auto acquiring. The article finds a new way to fulfill the task based on SIFT and BP neural network, and a special model is built. An experiment is done to testify the model. The result shows that the auto matching may be done, and the results are suitable to be used as a reference of the accuracy matching. The research would improve the efficiency to obtain the stereo keypoints and accelerate the measurement speed. Meanwhile, as an efficient technology, it will lend ideas to ensure the vehicle whose license is covered.