Abstract: This paper researches pre-obstacle detection in the case of blurred image caused by relative movement between the camera and the object, in accordance with the unstructured environment. This paper presents an improved two-dimensional entropy segmentation algorithm. Edge sharpness method is used to determine whether the image is a motion blurred image. If the answer is yes, the point spread function should be obtained using Fourier spectrum and Radon transform, and the image is restored by Lucy-Richardson method. In order to reduce the complexity of the obstacle detection algorithm, obstacles are coarsely positioned using the improved two-dimensional entropy segmentation algorithm, and the area threshold method  is used to eliminate noise. The pre-obstacle detection is completed. Simulation results demonstrate the feasibility and the effectiveness. This can provide reference for smart cars avoiding colliding with obstacles, and is the foundation for identifying obstacles.

Key words: obstacle detection, motion-blurred, image restoration, point spread function, 2-D entropy, area threshold, Radon transform

摘要： 针对在非结构化环境下，摄像机和拍摄目标相对运动造成的图像模糊情况，进行障碍物检测预处理。利用边缘锐度法判定获得的图像是否为运动模糊图像，如果是则利用傅里叶频谱和Radon变换得到点扩展函数后，利用 Lucy-Richardson法进行复原；为了降低后期障碍物检测的处理复杂度，利用提出的一种改进二维熵分割算法进行障碍物粗定位，面积阈值消除噪声影响，完成障碍物检测预处理。通过仿真实验验证了该方法的可行性和有效性，为智能汽车在非结构化环境下避障提供参考依据，为障碍物识别奠定基础。

关键词: 障碍物检测, 运动模糊, 图像复原, 点扩展函数, 二维熵, 面积阈值, Radon变换