低照度环境下非结构化道路区域识别

Unstructured road area recognition in low illumination environment

非结构化道路中道路区域与非道路区域的识别是自动驾驶车辆安全行驶和避障的重要课题。非结构化道路多存在于光照不足的环境,采集获得的图像质量差,从而影响对场景信息的提取。因此,本文提出了一种有效的低照度道路检测系统。首先,采用同态滤波算法对获取的低照度图像进行增强;其次,利用分割精度高的UNet语义分割网络得到包含道路区域的图像;最后,提出一种融合多种方法的边界拟合算法,拟合得出道路边界线。实验证明,该算法可以有效地在低照度环境下得到可行驶道路区域。

The identification of road area and non-road area in unstructured road is an important issue for safe driving and obstacle avoidance of autonomous vehicles. Unstructured roads mostly exist in the environment with insufficient light, and the acquired image quality is poor, which affects the extraction of scene information. Therefore, this paper proposes an effective road detection system for low illumination. Firstly, homomorphic filtering algorithm is used to enhance the low-illuminance image. Secondly, UNet semantic segmentation network with high segmentation accuracy is used to obtain images containing road regions. Finally, a boundary fitting method is proposed to obtain the road boundary. Experimental results show that the algorithm can effectively obtain the exercisable road region in low illumination environment.