光流法在机车安全行驶中的应用

在基于路况视频的机车安全行驶中,依据轨道与信号灯的相对位置先验知识可缩小信号灯检测范围,提高信号灯检测的效率和可靠性,但在某些复杂光照条件下(尤其是夜间)因检测不到轨道而无法提供信号灯区域信息。针对此问题,提出一种基于光流的信号灯区域预测方法,首先在感兴趣区域中使用KLT算子提取强角点,并采用基于金字塔的Lucas-Kanade光流法计算强角点光流;然后根据全局光流信息判断机车转向状态;最后结合转向状态与先验知识估计出信号灯区域。此外,还提出了利用转向状态进行轨道跟踪检错的方法,该方法可及时发现轨道跟踪错误并唤醒系统进行轨道重定位。实验表明,该方法能有效预测信号灯区域,使信号灯检测实时性更高,抗干扰能力更强。

Application of optical flow in locomotive safe driving

In the video-based locomotive safe driving assistant system, reducing the detecting region of signal light according to the location of rails and some prior knowledge can improve the efficiency and reliability of signal light detection. But it possibly fails to detect rail in some complicated illumination environments (especially in the night), and the signal light detecting region cannot be available. To solve this problem, this paper proposed an optical flow based approach for signal light region prediction. Firstly, it got the strong corners in region of interest by KLT operators, and calculated their optical-flow using Lucas-Kanade method with pyramids. After that it could get the locomotive turning state from the optical flow statistical information. In the end, it could dynamically estimate the signal light region according to the turning states and prior knowledge. Besides, it also proposed a method for rails false-tracking detecting by utilizing locomotive turning states, which could detect the tracking error timely and arouse the rails relocation. Experimental results demonstrate that the proposed method can estimate the signal region effectively and make the signal detection more real-time and more robust against disturbance.