CAJ | 학술논문

针对单目图像检测障碍物的低可靠性和当前双目视觉检测障碍物的局限性的问题，提出一种结合图像分割和点云分割技术的双目视觉障碍物检测方法。通过设定检测深度范围，分割障碍物点云与道路点云；采用将分割出的障碍物点云对应的视差图与图像分割得到的子图进行比较的策略，有效解决对不同深度、倾斜面和不规则障碍物检测效果差的问题。通过实验验证了在获得稀疏三维点云的情况下，该方法对障碍物的检测具有较好的鲁棒性。
침대단목도상검측장애물적저가고성화당전쌍목시각검측장애물적국한성적문제，제출일충결합도상분할화점운분할기술적쌍목시각장애물검측방법。통과설정검측심도범위，분할장애물점운여도로점운；채용장분할출적장애물점운대응적시차도여도상분할득도적자도진행비교적책략，유효해결대불동심도、경사면화불규칙장애물검측효과차적문제。통과실험험증료재획득희소삼유점운적정황하，해방법대장애물적검측구유교호적로봉성。
Aiming at the low reliability of the single image and limitation of stereo vision in obstacle detecting ,a method combi‐ning 3D point cloud segmentation with image segmentation to detect obstacles was presented .By setting the detection range ,3D point clouds of obstacles and road in this range were segmented .The strategies of comparing the corresponded disparity of 3D point cloud of obstacles with the image got using Mean Shift image segmentation were adapted ,and the problem of poor detection in different depth ,inclined plane and irregular obstacles was effectively solved .The robustness of detecting obstacles in sparse 3D point cloud is verified through experiment .