Embedded GPU zero-copy panoramic image splicing method and system based on improved L-ORB algorithm

An embedded GPU zero-copy panoramic image splicing method based on an improved L-ORB algorithm comprises the following steps: step 1, determining an overlapping region of a camera array by using a calibration file, and then extracting and matching feature points of images in the overlapping region by using the ORB algorithm; 2, performing exposure compensation on the original image based on brightness consistency prior; and step 3, GPU utilization is optimized by using a stream parallel strategy. The invention also comprises an embedded GPU zero-copy panoramic image splicing system based on the improved L-ORB algorithm. According to the method, the efficiency is improved, and GPU resources can be fully utilized. 一种改进L-ORB算法的嵌入式GPU零拷贝全景图像拼接方法，包括：步骤1，使用标定文件确定相机阵列的重叠区域，然后使用ORB算法对重叠区域的图像进行特征点提取和匹配；步骤2,基于亮度一致性先验对原始图像进行曝光补偿；步骤3,使用流并行策略优化GPU利用。本发明还包括一种改进L-ORB算法的嵌入式GPU零拷贝全景图像拼接系统。本发明提升了效率，并且可以充分利用GPU资源。