基于KCF-Hash-Match目标跟踪算法的高速列车车轮横向晃动识别方法

KCF-Hash-Match Target Tracking Algorithm for Identifying Wheel Lateral Sway of High-speed Train

良好的轮轨接触状态是高速列车安全运行的重要保障,轮轨相对横移量作为判别轮轨接触状态的重要指标,直接决定着车轮是否存在脱轨危险,基于机器视觉的轮轨在线跟踪监测技术可为高速列车轮轨接触点状态识别提供新的思路。测试采用在转向架底部安装摄像机的测试方法,获取列车高速运行时的轮轨接触图像。综合考虑核相关滤波算法跟踪速度快和模板匹配算法跟踪精度高的优点,并融入哈希算法原理。提出一种能够识别高速列车轮轨接触几何状态的核相关滤波-哈希-模板匹配算法(Kernel correlation filtering-Hash-Match,KCF-Hash-Match),用于复杂运营环境下高速列车车轮横向异常晃动的监测。结果表明,KCF-Hash-Match算法能够解决传统核相关滤波算法在跟踪快速移动目标时存在的误差积累问题,当允许的误差阈值为8个像素点时,准确率达到99.8%;与现场实测数据对比表明,KCF-Hash-Match算法具有较高的跟踪精度,可实现高速列车车轮横向晃动的安全监测。

A good wheel-rail contact state is an important guarantee for the running safety of high-speed trains.The relative lateral movement of wheel and rail is an important indicator for judging the wheel-rail contact state,which directly determines the danger of wheel derailment.The wheel-rail online tracking monitoring technology based on machine vision can provide a new idea for the state identification of the wheel-rail contact point of high-speed trains.The test adopts the method of installing a camera at the bottom of the bogie to obtain images of wheel rail contact during high-speed train operation.Considering the advantages of fast tracking speed of the KCF algorithm and the high tracking accuracy of the template matching algorithm,and incorporating the principle of hashing algorithm,a Kernel Correlation Filtering-Hash-Match algorithm(KCF-Hash-Match)that can automatically identify wheel-rail contact geometry is proposed,which is applied to the monitoring of abnormal wheel lateral sway of high-speed trains under complex condition.The results show that the KCF-Hash-Match algorithm can solve the error accumulation problem existing in the traditional KCF algorithm when deal with tracking fast moving targets.When the allowable error threshold is 8 pixels,the accuracy rate can attain 99.8%.Compared with the field measured data,the KCF-Hash-Match algorithm has high tracking accuracy and can realize the safety monitoring of high-speed train wheel lateral sway behavior.