基于Yolov3的自动驾驶目标检测

Automatic Driving Target Detection Based on Yolov3

自动驾驶场景下,为解决前方障碍物的检测实时性与检测准确度难以权衡的问题,采用空间金字塔的思想,对Yolov3主网络部分进行修改,形成了Yolov3-bt目标检测算法,提高了目标检测准确率。通过调整输入图片分辨率,加快了检测速度,从而得到了检测时间与检测准确率的最优匹配值。最后,通过实验验证结果表明,改进后的Yolov3-bt目标检测算法可兼顾前方障碍物检测实时性及检测准确度。Yolov3-bt-416平均处理1张图片约耗时29.89 ms,前方障碍物识别准确率在IOU值为0.6的前提下,行人检测率可达87%,交通类障碍物可达92%。

Under the automatic driving scene,it is difficult to balance the reduction of detection time and the improvement of detection accuracy.In order to solve this problem,based on the idea of spatial pyramid,the Yolov3 BT target detection algorithm is formed by modifying the main network of Yolov3,which improves the accuracy of target detection.By adjusting the resolution of the input image,the detection speed is increased.Experiments show that:Yolov3-bt-416 processes a picture within 29.89 ms.When the IOU is 0.6,the accuracy of detecting pedestrians is 87%,and the accuracy of detecting traffic obstacles is 92%.Therefore,the detection time is accelerated while ensuring the detection accuracy.