为提高井下复杂恶劣环境下刮板输送机直线状态监测精度和实时性，提出了一种基于统一计算设备架构（CUDA）加速动态规划优化全景拼接的刮板输送机直线状态监测方法。首先，同步2路摄像头获取的刮板输送机图像的帧数、分辨率参数，对输入的视频流进行暗通道清晰化处理，以消除井下煤尘、水雾等的干扰。其次，使用ORB算法检测和计算2路视频帧的特征点和描述子，通过K最近邻（KNN）匹配计算特征点间的匹配对，利用设置阈值比例的方法过滤错误的匹配点，使用随机抽样一致（RANSAC）算法计算出用于图像透视变换的单应性矩阵。然后，基于CUDA将读取Sobel算子、计算梯度、计算总能量差异、循环初始化权重与路径、寻找最佳接缝线分配到不同的线程中，并定义计算能量图和寻找最佳接缝线的核函数，完成2路图像沿接缝线融合的全景拼接。最后，使用霍夫变换方法对全景拼接的刮板输送机图像中部槽挡煤板进行直线拟合，并将拟合的直线绘制在全景拼接图像上，以反映刮板输送机的直线状态。实验及测试结果表明，CUDA加速动态规划优化全景拼接痕迹不明显，且处理速度快；通过霍夫变换对中部槽挡煤板拟合的直线与刮板输送机直线具有较好的一致性，可用于刮板输送机直线状态监测。

To enhance the accuracy and real-time performance of straightness monitoring for scraper conveyors in complex and harsh underground environments, a straightness monitoring method for scraper conveyors based on compute unified device architecture (CUDA)-accelerated dynamic programming and optimized panoramic stitching was proposed. First, the frame rates and resolution parameters of images captured by two synchronized cameras were aligned, and the input video stream underwent dark channel enhancement to eliminate the interference of coal dust, water mist, and other factors underground. Then, the oriented FAST and rotated BRIEF (ORB) algorithm was used to detect and calculate feature points and descriptors from the two video frames. Feature point matches were calculated using K-nearest neighbors (KNN), with incorrect matches filtered using a threshold ratio. A homography matrix for image perspective transformation was calculated using the random sample consensus (RANSAC) algorithm. CUDA was employed to accelerate the processing by assigning tasks such as Sobel operator reading, gradient computation, total energy difference calculation, loop initialization of weights and paths, and optimal seam finding to different threads. Kernel functions for energy map computation and seamline optimization were defined, enabling the seamless fusion of the two images along the seamline to complete panoramic stitching. Finally, the Hough transform was applied to perform linear fitting of the coal blocking plate in the middle trough of the stitched panoramic image. The fitted line was superimposed on the panoramic image to reflect the straightness status of the scraper conveyor. The experiment and test results showed that the CUDA-accelerated dynamic programming significantly reduced visible stitching artifacts while ensuring high processing speed. The straight line fitted by the Hough transform closely matched the actual straightness of the scraper conveyor, demonstrating its effectiveness for straightness monitoring of scraper conveyors.