基于粒子滤波的机车信号灯跟踪方法

Particle filter based railway traffic lights tracking

为保障机车行驶安全,由车载高清摄像机获取路况视频并识别信号灯及其颜色状态时,视频中信号灯目标尺度变化大、机车行驶抖动、复杂光环境及光圈自适应调节滞后等因素使得信号灯鲁棒跟踪与识别具有不小难度.针对信号灯跟踪问题,本文提出一种带检测矫正的粒子滤波跟踪方法,该方法在粒子滤波框架下对信号灯进行跟踪,并通过一个在线更新的模板对滤波结果进行检测矫正,以提高跟踪结果的准确性.为提高跟踪算法对光照以及目标尺度变化的适应能力,本文在对信号灯建模时融合了HSV颜色特征与局部二元模式特征.实验结果表明,该方法在较复杂的场景下能够很好地对信号灯进行实时鲁棒的跟踪,并且跟踪结果具有较高的准确性.

For insuring the railway locomotive driving safety,the railway traffic light and its color state can be recognized from high-definition road condition video that acquired by on-vehicle camera.However,big scale change of traffic lights,locomotive shaking,complex lighting condition,and the delay of aperture adjusting would make the traffic light tracking and recognition difficult.In this paper,the authors will focus on the railway traffic light tracking and put forward a detect-rectified Particle Filter（PF） tracking approach,which first tracks traffic light with particle filter,and then applies an online-updating template to improve the tracking accuracy.In addition,aim to increase the tracking adaptability,the authors model the traffic light by combining HSV color feature and Local Binary Pattern（LBP） feature.Experimental results demonstrate that the proposed method can effectively track the traffic light in complicated background in real-time with good robustness and high locating accuracy.