结合多尺度特征的改进YOLOv2车辆实时检测算法

被引:10
作者
金宇尘
罗娜
机构
[1] 华东理工大学化工过程先进控制和优化技术教育部重点实验室
关键词
无人驾驶系统; 车辆目标; 实时检测; 深度学习; 难样本生成;
D O I
10.16208/j.issn1000-7024.2019.05.047
中图分类号
TP391.41 []; TP18 [人工智能理论];
学科分类号
080203 ; 081104 ; 0812 ; 0835 ; 1405 ;
摘要
为解决车辆目标检测过程中较小车辆及被遮挡车辆容易被漏检、错检的问题,基于深度学习先进算法YOLOv2,提出一种结合多尺度特征的改进车辆目标实时检测方法。通过融合多尺度特征层信息建立额外检测特征层对大小不同的车辆进行检测;构建自适应损失函数作为置信度目标函数解决正负样本数量不平衡问题,提高置信度预测准确性;开发一种新的难样本生成方法对网络进行训练,减小错检发生的概率。实验结果表明,该方法在运行速度满足实时检测要求的情况下,能显著降低车辆目标漏检、错检率,平均准确率提高9%以上。
引用
收藏
页码:1457 / 1463+1476 +1476
页数:8
相关论文
共 19 条
  • [1] Accurate Single Stage Detector Using Recurrent Rolling Convolution. REN J,CHEN X,LIU J et al. IEEE Conference on Computer Vision and Pattern Recognition . 2017
  • [2] Focal Loss for Dense Object Detection. Lin T.Y,Goyal P,Girshick R.et al. IEEEInternational Conference on Computer Vision (ICCV) . 2017
  • [3] SSD:Single Shot MultiBox Detector. Liu W,Anguelov D,Erhan D,et al. European Conference on Computer Vision . 2016
  • [4] Feature pyramid networks for object detection. Lin T Y,Dollar P,Girshick R.et al. Computer Vision and Pattern Recognition . 2017
  • [5] Imagenet large scale visual recognition challenge. Russakovsky O,Deng J,Su H,et al. International Journal of Computer Vision . 2014
  • [6] YOLO9000:Better,Faster,Stronger. Redmon,J,Farhadi,A. 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) . 2017
  • [7] You only lookonce:unified,real-time object detection. Redmon J,Divvala S,Girshick R,et a1. Proceedingsof the 2016 Conference on Computer Vision and PatternRecognition . 2016
  • [8] Faster R-CNN:Towards Real-Time Object Detection with Region Proposal Networks. Ren S,Girshick R,Girshick R,et al. IEEE Transactions on Pattern Analysis&Machine Intelligence . 2015
  • [9] Spatial pyramid pooling in deep convolutional networks for visual recognition. HE K,ZHANG X,REN S,et al. IEEE Transactions on Pattern Analysis.&.Machine Intelligence . 2014
  • [10] Fast R-CNN. Girshick R. IEEE International Conference on Computer Vision . 2015