基于语义分割的近距离全向车距检测方法研究

Research on Close-range Omnidirectional Vehicle Distance Detection Method Based on Semantic Segmentation

为获取主车与周围近距离车辆之间的距离,确保驾驶安全以及预防剐蹭事故,提出了一种基于视觉的近距离全向车距检测方法。首先,在车身四周安装四路鱼眼镜头,使用语义分割模型对采集到的图像中的车辆进行语义分割;然后将车辆分割后的四路图像拼接为360°环视图像;根据环视图像中像素与实际地面距离的比例关系,通过计算图像中周围车辆外缘到主车外缘的最近距离,实现近距离全向车距的检测。实验结果表明:该方法检测的车距平均误差为12.6%,且误差在15%以内的占比超过90%。

In order to obtain the distance between the main vehicle and the surrounding close-range vehicles to ensure driving safety as well as to prevent cut-and-run accidents,a vision-based close-range omni-directional vehicle distance de-tection method is proposed.Firstly,a four-way fisheye lens is installed around the vehicle body,and a semantic segmenta-tion model is used to semantically segment the vehicles in the captured image;then the four-way image after vehicle seg-mentation is stitched into a 360-degree ring-view image;according to the proportionality between the pixels in the ring-view image and the distance from the actual ground,the detection of the proximity omni-directional distance is realized by calculating the closest distance from the outer edges of the surrounding vehicles to the outer edge of the main vehicle in the image.The experimental results show that the average error of the close-range distance detection proposed in this paper is 12.6%,and the percentage of error within 15%is more than 90%.