Underground pedestrian detection model based on Dense-YOLO network
-
摘要: 行人检测是实现矿用车辆无人化的一项关键技术，煤矿井下弱光环境中捕获的图像可见度不佳，极大地影响了行人检测效果。现有行人检测方法忽略了井下弱光环境对目标检测精度的影响，检测效果不理想。针对该问题，提出一种基于Dense-YOLO网络的井下行人检测模型。将弱光图像分解为光照图和反射图，采用Gamma变换、加权对数变换、限制对比度的自适应直方图均衡（CLAHE）对光照图进行增强处理，采用亮度权值和色彩权值对增强后的图像进行加权融合；采用双边滤波算法对反射图进行处理，以增强图像纹理；将增强后的光照图和经过双边滤波处理的反射图逐点相乘，重构出RGB图，并采用ROF去噪模型对融合后的图像进行全局去噪，得到最终的增强图像。将含有残差块的Dense模块添加到YOLOv3中，构建基于Dense-YOLO网络的井下行人检测模型，残差块的加入有利于避免在网络训练过程中出现梯度消失和梯度爆炸等问题。实验结果表明：对弱光图像进行增强处理能够有效提高图像可见度和行人检测效果；Dense-YOLO网络对增强图像的漏检率为4.55%，相较于RetinaNet网络降低了14.91%，基于Dense-YOLO网络的井下行人检测模型有效降低了行人检测漏检率。
-
关键词:
- 井下行人检测 /
- 弱光图像增强 /
- Dense-YOLO /
- YOLOv3 /
- Gamma变换 /
- 加权对数变换 /
- 限制对比度的自适应直方图均衡
Abstract: The pedestrian detection is a key technology to realize unmanned mining vehicles. The visibility of images captured in low light environment in coal mine is poor, which greatly affects the effect of pedestrian detection. The existing pedestrian detection methods ignore the influence of underground low light environment on target detection precision, and the detection effect is not ideal. In order to solve this problem, an underground pedestrian detection model based on Dense-YOLO network is proposed. The low light images are decomposed into light image and reflection image, and the light image is enhanced by Gamma transformation, weighted logarithmic transformation and contrast-limited adaptive histogram equalization (CLAHE). The enhanced images are weighted and fused by brightness weight and color weight. The bilateral filtering algorithm is used to process the reflection image to enhance the texture of the image. The enhanced light image and the reflection image processed by bilateral filtering are multiplied point by point to reconstruct the RGB image, and the ROF denoising model is used to denoise the fused image globally to obtain the final enhanced image. The dense module with residual block is added to YOLOv3 to build underground pedestrian detection model based on Dense-YOLO network. The addition of residual block is beneficial to avoid gradient disappearance and gradient explosion in the network training process. The experimental results show that the image visibility and pedestrian detection can be improved effectively by enhancing the low light image. The missed detection rate of Dense-YOLO network for enhanced images is 4.55%, which is 14.91% lower than that of RetinaNet network. The underground pedestrian detection model based on Dense-YOLO network effectively reduces the missed detection rate of pedestrian detection. -
-
![]()
图 3 基于Dense-YOLO网络的井下行人检测模型
Figure 3. Underground pedestrian detection model based on Dense-YOLO network
![]()
图 4 弱光环境下行人检测结果对比
Figure 4. Comparison of pedestrian detection results in low light environments
表 1 RetinaNet网络和Dense-YOLO网络检测结果
Table 1 Detection results of RetinaNet network and Dense-YOLO network
图像 网络 漏检
率/%mAP/% 运行
时间/ms弱光
图像RetinaNet 96.15 3.42 50.49 Dense-YOLO 60.81 31.24 51.34 增强
图像RetinaNet 19.46 92.63 49.55 Dense-YOLO 4.55 87.79 50.27 
下载: 导出CSV
-
[1] 张小艳,郭海涛. 基于改进混合高斯模型的井下目标检测算法[J]. 工矿自动化,2021,47(4):67-72. ZHANG Xiaoyan,GUO Haitao. Underground target detection algorithm based on improved Gaussian mixture model[J]. Industry and Mine Automation,2021,47(4):67-72.
[2] 董昕宇,师杰,张国英. 基于参数轻量化的井下人体实时检测算法[J]. 工矿自动化,2021,47(6):71-78. DONG Xinyu,SHI Jie,ZHANG Guoying. Real-time detection algorithm of underground human body based on lightweight parameters[J]. Industry and Mine Automation,2021,47(6):71-78.
[3] 谢林江,季桂树,彭清,等. 改进的卷积神经网络在行人检测中的应用[J]. 计算机科学与探索,2018,12(5):708-718. DOI: 10.3778/j.issn.1673-9418.1708030 XIE Linjiang,JI Guishu,PENG Qing,et al. Application of preprocessing convolutional neural network in pedestrian detection[J]. Journal of Frontiers of Computer Science and Technology,2018,12(5):708-718. DOI: 10.3778/j.issn.1673-9418.1708030
[4] 刘备战,赵洪辉,周李兵. 面向无人驾驶的井下行人检测方法[J]. 工矿自动化,2021,47(9):113-117. LIU Beizhan,ZHAO Honghui,ZHOU Libing. Unmanned driving-oriented underground mine pedestrian detection method[J]. Industry and Mine Automation,2021,47(9):113-117.
[5] LIN T Y,P GOYAL P,GIRSHICK R,et al. Focal loss for dense object detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,2020,42(2):318-327. DOI: 10.1109/TPAMI.2018.2858826
[6] HE Kaiming,SUN Jian,TANG Xiaoou,et al. Single image haze removal using dark channel prior[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,2011,33(12):2341-2353. DOI: 10.1109/TPAMI.2010.168
[7] FU Xueyang,ZENG Delu,HUANG Yue,et al. A fusion-based enhancing method for weakly illuminated images[J]. Signal Processing,2016,129:82-96. DOI: 10.1016/j.sigpro.2016.05.031
[8] WANG Shuhang,ZHENG Jin,HU Haimiao,et al. Naturalness preserved enhancement algorithm for non-uniform illumination images[J]. IEEE Transactions on Image Processing,2013,22(9):3538-3548. DOI: 10.1109/TIP.2013.2261309
[9] REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: unified, real-time object detection[EB/OL]. （2016-05-09） [2021-10-10]. https://arxiv.org/pdf/1506.02640.pdf.
[10] REDMON J, FARHADI A. YOLOv3: an incremental improvement[EB/OL]. （ 2018-04-08） [2021-10-10]. https://arxiv.org/pdf/1804.02767.pdf.
[11] HUANG Gao, LIU Zhuang, MAATEN L V D, et al. Densely connected convolutional networks[EB/OL]. （ 2018-01-28）[2021-10-10]. https://arxiv.org/pdf/1608.06993.pdf.
[12] LIN T Y, MAIRE M, BELONGIE S, et al. Microsoft COCO: common objects in context[EB/OL]. （2015-02-21）[2021-10-10]. https://arxiv.org/pdf/1405.0312.pdf.
-
期刊类型引用(2)
1. 程林. 电子定量包装秤自动化二次校准系统设计. 现代电子技术. 2023(24): 34-38 . 
百度学术
2. 胡飞，陈彩蓉，葛飞，曹文韬，於海明. 生鲜乳快速冷却装置监测系统设计. 农业工程. 2022(09): 41-44 . 百度学术
其他类型引用(4)
计量
- 文章访问数: 235
- HTML全文浏览量: 138
- PDF下载量: 53
- 被引次数: 6
