基于Faster R-CNN的高速公路抛落物检测

采用传统的基于图像处理的检测方法对高速公路抛落物进行检测不仅耗时耗力,而且检测效果不理想,为解决该问题,提出一种基于Faster R-CNN的深度学习检测方法.在原始Faster R-CNN的基础上,采用残差网络Resnet101代替传统的VGG-16网络和ZFNet网络,作为图像特征提取网络;采用尺寸为4像素、8像素和16像素的锚框代替原始锚框,得到高速公路抛落物检测模型.采用自制的高速公路抛落物数据集对该检测方法的有效性进行验证,结果显示,采用该方法检测的平均准确率达到了91.75％,相比原始的Faster R-CNN算法和yolov3算法,分别提高了7.02％和11.13％.

Detection of Dropped Objects on Highway with Faster R-CNN

The traditional detection method based on image processing has been found not to be suited to detect dropped objects.A deep learning detection method based on Faster R-CNN is introduced.The residual network Resnet101 is used as the image feature extraction network instead of VGG-16/ZFNet network as in normal rcnns.The anchor frame sizes are also changed into 4,8 and 16.In this way,the highway dropped object detection model is constructed.The detection algorithm is verified.It is demonstrated that an average accuracy rate of 91.75%is achieved,7.02%/11.13%higher than that with original Faster R-CNN algorithm/yolov3 algorithm.