基于帧间差分累积的铁路限界异物检测提取算法

为满足轨道交通全自动运行系统对铁路限界内异物检测提取的需求,改进帧间差分法,提出一种以帧间差分累积为基础的铁路限界内异物检测提取算法。算法针对图像序列帧匹配提取的轨道线为依据标定限界区域,通过多帧隔帧帧差法得到差分结果,根据铁路限界内道床纹理特征,通过数学形态学实现背景纹理的重构来降低背景噪声影响。最后,以侧向差分灰度的累积投影值来动态确定不同环境下的异物前景范围,并通过最大类间方差法提取得到前景目标。通过对47个路轨场景进行测试,算法对有前景目标场景的目标检测率为96.87%,定位提取过程的平均耗时为137 ms。实验结果表明:算法可完成对运动背景下的轨道限界内前景目标的定位和提取,具有较好的实时性和准确性。

Detection and Extraction Algorithm of Foreign Object in Railway Clearance Based on Inter Frame Difference Accumulation

Aiming at the requirement of foreign object detection and extraction in railway clearance of rail transit full-automatic operation system, the inter-frame difference method is improved, and an algorithm for detecting and extracting foreign object in railway clearance based on inter-frame difference accumulation is proposed. The algorithm calibrates the boundary area based on the matching extracted track line for the image sequence frame, and obtains the difference results by means of the frame difference method of multi-frame interlacing. According to the texture features of the ballast bed in the railway clearance, the background texture is reconstructed by mathematical morphology to reduce the background noise effect. Finally, the projected range of foreign object in different environments is dynamically determined by the cumulative projection value of the lateral difference grayscale. By testing 47 track scenes, the target detection rate of the algorithm is 96.87% for foreground target scenes, and the average time spent in locating and extracting is 137 ms. The experimental results show that the algorithm can locate and extract the foreground target in the railway clearance under the moving background, and has preferable real-time performance and accuracy.