摘 要：针对实际交通环境下行人和车辆检测问题，提出一种基于 YOLOv3 改进的目标检测网络 YOLO-CP，对 YOLOv3网络结构进行压缩剪枝，并进行特征提取的优化，使用自主采集标注的交通数据集，进行稀疏化训练。在实际交通场景中，YOLO-CP 在 GPU 下检测速度达到 25 帧 / 秒，车辆检测准确率达到 96.0%，行人检测准确率达到 93.3%，优化算法满足了ADAS 对实时性和高精度的要求。
关键词：行人检测；车辆检测；YOLOv3；ADAS
DOI:10.19850/j.cnki.2096-4706.2021.07.015
基金项目：2020 年山东华宇工学院科技计 划项目（2020KJ16）
中图分类号：TP391.41；TP18 文献标识码：A 文章编号：2096-4706（2021）07-0059-04
Research on Vehicle and Pedestrian Detection Algorithm Based on Deep Learning
ZHANG Feng
（Shandong Huayu University of Technology，Dezhou 253034，China）
Abstract：Aiming at the problem of pedestrian and vehicle detection in actual traffic environment，this paper proposes an improved target detection network YOLO-CP based on YOLOv3，which compresses and prunes the YOLOv3 network structure， optimizes the feature extraction，and uses the independently collected and labeled traffic data set for sparse training. In the actual traffic scene，the detection speed of YOLO-CP under the GPU reaches 25 frames/s，the vehicle detection accuracy rate reaches 96.0%，and the pedestrian detection accuracy rate reaches 93.3%.The optimization algorithm meets the real-time and high-precision requirements of ADAS.
Keywords：pedestrian detection；vehicle detection；YOLOv3；ADAS
参考文献：
[1] ROSS G，DONAHUE J，DARRELL T，et al. Rich feature hierarchies for accurate object detection and semantic segmentation [C]//2014 IEEE Conference on Computer Vision and Pattern Recognition.Columbus：IEEE，2014：580-587.
[2] GIRSHICK R. FAST R-CNN [C]//2015 IEEE International Conference on Computer Vision （ICCV）.Santiago：IEEE，2015： 10-15.
[3] REN S Q，HE K M，Girshick R，et al. Faster R-CNN： Towards Real-Time Object Detection with Region Proposal Networks [J]. IEEE transactions on pattern analysis and machine intelligence，2017， 39（6）：1137-1149.
[4] REDMON J，DIVVALA S，GIRSHICK R，et al. You Only Look Once：Unified，Real-Time Object Detection [C]//2016 IEEE Conference on Computer Vision and Pattern Recognition（CVPR）.Las Vegas：IEEE，2016：13.
[5] REDMON J，FARHADI A. YOLO9000：Better，Faster， Stronger [C]//2017 IEEE Conference on Computer Vision and Pattern Recognition （CVPR）.Honolulu：IEEE，2017：6517-6525.
[6] LIU W， ANGUELOV D， ERHAN D，et al. SSD： Single Shot MultiBox Detector [C]//Computer Vision-ECCV 2016. Amsterdam：Springer，Cham，2016：21-37.
[7] REDMON J，FARHADI A. YOLOv3：An Incremental Improvement [J/OL].arXiv：1804.02767 [cs.CV].（2018-04-18）. https://arxiv.org/abs/1804.02767.
[8] HE K M，ZHANG X Y，REN S Q，et al. Deep Residual Learning for Image Recognition [C]//2016 IEEE Conference on Computer Vision and Pattern Recognition （CVPR）.Las Vegas： IEEE，2016：770-778.
[9] REDMON J. Darknet：Darknet：Open Source Neural Networks in C. [EB/OL].[2021-03-10].http://pjreddie.com/darknet/.
[10] IOFFE S，SZEGEDY C. Batch Normalization：Accelerating Deep Network Training by Reducing Internal Covariate Shift [J/ OL].arXiv：1502.03167 [cs.LG].（2015-02-11）.https://arxiv.org/ abs/1502.03167.
[11] LIN T Y，Dollár P，Girshick R，et al. Feature Pyramid Networks for Object Detection [C]//2017 IEEE Conference on Computer Vision and Pattern Recognition（CVPR）.Honolulu：IEEE，2017： 936-944.
[12] HAN S，MAO H，DALLY W J. Deep Compression： Compressing Deep Neural Networks with Pruning，Trained Quantization and Huffman Coding [J/OL].arXiv：1510.00149 [cs.CV]. （2015-10-01）.https://arxiv.org/abs/1510.00149.
作者简介：张凤（1991—），女，汉族，山东临沂人，讲师， 硕士研究生，研究方向：图像处理。