东北大学学报(自然科学版) ›› 2025, Vol. 46 ›› Issue (5): 29-36.DOI: 10.12068/j.issn.1005-3026.2025.20230297
收稿日期:
2023-10-27
出版日期:
2025-05-15
发布日期:
2025-08-07
通讯作者:
车德福
作者简介:
艾散·西尔艾力(1997—),男,新疆尉犁人,东北大学硕士研究生基金资助:
Aisan XIERAILI1, De-fu CHE1(), Duo WANG2, Tian YU3
Received:
2023-10-27
Online:
2025-05-15
Published:
2025-08-07
Contact:
De-fu CHE
摘要:
基于机器视觉的环境感知技术是智慧交通领域的关键任务之一.传统深度学习算法通常只能满足单一场景下的个别目标检测任务,难以应对复杂交通环境下的智能感知需求.为提高车辆在复杂环境下的智能感知能力,提出了一种改进的YOLOv8目标检测网络模型,结合注意力机制、优化器和可变形卷积层,实现了在复杂城市交通环境下的多目标检测.采用YOLOv4,YOLOv8及改进的YOLOv8算法对复杂交通环境样本图进行目标检测对比实验.结果表明,与YOLOv4,YOLOv8相比,改进的YOLOv8算法的平均精度分别提高了40.76%和16.92%.该算法的检测准确性与实时性满足实际应用需求,可通过多传感器信息融合,实现在复杂城市交通环境下的智能感知.
中图分类号:
艾散·西尔艾力, 车德福, 王夺, 喻甜. 基于目标检测的复杂城市交通环境感知技术及应用[J]. 东北大学学报(自然科学版), 2025, 46(5): 29-36.
Aisan XIERAILI, De-fu CHE, Duo WANG, Tian YU. Perception Technology and Application of Complex Urban Traffic Environment Based on Target Detection[J]. Journal of Northeastern University(Natural Science), 2025, 46(5): 29-36.
图5 标准卷积和可变形卷积的采样位置示意图(a)—标准卷积规则采样网格; (b)—可变形卷积采样网格(带偏移量);(c)—可变形卷积采样网格(尺度变换); (d)—可变形卷积(旋转变换).
Fig.5 Schematic diagram of sampling positions of standard convolution and deformable convolution
训练参数 | 数值 | 训练参数 | 数值 |
---|---|---|---|
epoch | 100 | imgsize | 640 |
batch | 16 | workers | 4 |
optimizer | Adam | patience | 50 |
save_period | 10 | close_mosaic | 10 |
表1 训练参数设置
Table 1 Training parameter setting
训练参数 | 数值 | 训练参数 | 数值 |
---|---|---|---|
epoch | 100 | imgsize | 640 |
batch | 16 | workers | 4 |
optimizer | Adam | patience | 50 |
save_period | 10 | close_mosaic | 10 |
测试图像 | 人工标记图像 | 本文模型检测结果 |
---|---|---|
![]() | ![]() | ![]() |
![]() | ![]() | ![]() |
表2 实验结果
Table 2 Experimental results
测试图像 | 人工标记图像 | 本文模型检测结果 |
---|---|---|
![]() | ![]() | ![]() |
![]() | ![]() | ![]() |
模型 | 主干网络 | 输入尺寸/像素 | mAP/% | FPS | IoU |
---|---|---|---|---|---|
YOLOv4 | CSPDarknet53 | 416×416 | 44.19 | 47 | 0.5 |
YOLOv8 | CSPDarknet+C2F | 640×640 | 53.20 | 87 | 0.5 |
改进YOLOv8 | CSPDarknet+C2F | 640×640 | 62.20 | 78 | 0.5 |
表3 不同网络模型的性能对比
Table 3 Performance comparison of different network models
模型 | 主干网络 | 输入尺寸/像素 | mAP/% | FPS | IoU |
---|---|---|---|---|---|
YOLOv4 | CSPDarknet53 | 416×416 | 44.19 | 47 | 0.5 |
YOLOv8 | CSPDarknet+C2F | 640×640 | 53.20 | 87 | 0.5 |
改进YOLOv8 | CSPDarknet+C2F | 640×640 | 62.20 | 78 | 0.5 |
目标类型 | 模型平均精度mAP50 | 改进模型精度提升量 | |||
---|---|---|---|---|---|
YOLOv4 | YOLOv8 | 改进YOLOv8 | 相对YOLOv4 | 相对YOLOv8 | |
总平均 | 44.19 | 53.20 | 62.20 | 18.01 | 9.00 |
交通标志牌 | 55.59 | 68.40 | 68.30 | 12.71 | -0.10 |
路面破损 | 57.60 | 73.60 | 81.10 | 23.50 | 7.50 |
井盖 | 50.95 | 47.50 | 51.10 | 0.15 | 3.60 |
井盖丢失 | 96.62 | 86.50 | 97.40 | 0.78 | 10.90 |
垃圾桶 | 53.95 | 71.80 | 75.80 | 21.85 | 4.00 |
垃圾桶倾倒 | 56.93 | 95.00 | 99.50 | 42.57 | 4.50 |
雨水立箅 | 53.26 | 58.80 | 67.40 | 14.14 | 8.60 |
雨水立箅破损 | 00.00 | 33.20 | 99.50 | 99.50 | 66.30 |
火焰 | 45.07 | 63.10 | 65.40 | 20.33 | 2.30 |
违规摆摊 | 44.19 | 47.00 | 54.70 | 10.51 | 7.70 |
表4 实验结果对比 (%)
Table 4 Comparison of experimental results
目标类型 | 模型平均精度mAP50 | 改进模型精度提升量 | |||
---|---|---|---|---|---|
YOLOv4 | YOLOv8 | 改进YOLOv8 | 相对YOLOv4 | 相对YOLOv8 | |
总平均 | 44.19 | 53.20 | 62.20 | 18.01 | 9.00 |
交通标志牌 | 55.59 | 68.40 | 68.30 | 12.71 | -0.10 |
路面破损 | 57.60 | 73.60 | 81.10 | 23.50 | 7.50 |
井盖 | 50.95 | 47.50 | 51.10 | 0.15 | 3.60 |
井盖丢失 | 96.62 | 86.50 | 97.40 | 0.78 | 10.90 |
垃圾桶 | 53.95 | 71.80 | 75.80 | 21.85 | 4.00 |
垃圾桶倾倒 | 56.93 | 95.00 | 99.50 | 42.57 | 4.50 |
雨水立箅 | 53.26 | 58.80 | 67.40 | 14.14 | 8.60 |
雨水立箅破损 | 00.00 | 33.20 | 99.50 | 99.50 | 66.30 |
火焰 | 45.07 | 63.10 | 65.40 | 20.33 | 2.30 |
违规摆摊 | 44.19 | 47.00 | 54.70 | 10.51 | 7.70 |
[1] | 高德芝,段建民,郑榜贵,等.智能车辆环境感知传感器的应用现状[J]. 现代电子技术, 2008(19): 151-156. |
Gao De-zhi, Duan Jian-min, Zheng Bang-gui, et al. Application status of intelligent vehicle environmental sensing sensor [J]. Modern Electronic Technology, 2008(19): 151-156. | |
[2] | Wang K, Gou C, Zheng N, et al. Parallel vision for perception and understanding of complex scenes: methods, framework, and perspectives[J]. Artificial Intelligence Review, 2017, 48(3): 299-329. |
[3] | Aufrère R, Gowdy J, Mertz C, et al. Perception for collision avoidance and autonomous driving [J]. Mechatronics, 2003, 13(10): 1149-1161. |
[4] | Montemerlo M, Becker J, Bhat S, et al. Junior: the Stanford entry in the urban challenge [J]. Journal of Field Robotics, 2008, 25(9): 569-597. |
[5] | Leonard J, How J, Teller S, et al. A perception-driven autonomous urban vehicle[J]. Journal of Field Robotics, 2008, 25(10): 727-774. |
[6] | 谢志萍,雷莉萍.智能网联汽车环境感知技术的发展和研究现状[J]. 成都工业学院学报, 2016, 19(4): 87-92. |
Xie Zhi-ping, Lei Li-ping. Development and research status of intelligent networked automotive environment awareness technology[J]. Journal of Chengdu Technological University, 2016, 19(4): 87-92. | |
[7] | Redmon J, Divvala S, Girshick R, et al. You only look once: unified, real-time object detection[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas, 2016:779-788. |
[8] | Asmaa B, Khalid Z. Optimizing CNN-BiGRU performance: mish activation and comparative analysis[J]. International Journal of Computer Networks & Communications, 2024, 16(3): 69-87. |
[9] | Wang Y F, Hua C C, Ding W L, et al. Real-time detection of flame and smoke using an improved YOLOv4 network[J]. Signal, Image and Video Processing, 2022, 16(4): 1-8. |
[10] | 张凯祥,朱明.基于YOLOv5的多任务自动驾驶环境感知算法[J]. 计算机系统应用, 2022, 31(9): 226-232. |
Zhang Kai-xiang, Zhu Ming. Multi-task automatic driving environment perception algorithm based on YOLOv5 [J]. Computer Systems & Applications, 2022, 31(9): 226-232. | |
[11] | Fei X, Li T H, Xiao Y G, et al. Research on YOLOv3 model compression strategy for UAV deployment[J]. Cognitive Robotics, 2024, 4: 8-18. |
[12] | Guo K Y, Cheng B H, Min Y, et al. A pavement distresses identification method optimized for YOLOv5s[J]. Scientific Reports, 2022, 12(1): 1-15. |
[13] | 郭振宇,高国飞.基于YOLO v4的复杂路口下人车混行检测算法研究[J]. 信息技术与信息化, 2021(2): 236-240. |
Guo Zhen-yu, Gao Guo-fei. Research on detection algorithm of mixed traffic between people and vehicles at complex intersections based on YOLO v4 [J]. Information Technology and Informatization, 2021(2): 236-240. | |
[14] | Łysakowski M, Żywanowski K, Banaszczyk A, et al. Real-time onboard object detection for augmented reality: enhancing head-mounted display with YOLOv8[C]//IEEE International Conference on Edge Computing and Communications. Chicago, 2023:364-371. |
[15] | Vaswani A, Shazeer N, Parmar N, et al. Attention is all you need[J]. Advances in Neural Information Processing Systems, 2017, 30: 6000-6010. |
[16] | 李鸿,邹俊颖,谭茜成,等.面向医学图像分割的多注意力融合网络[J]. 计算机应用, 2022, 42(12): 3891-3899. |
Li Hong, Zou Jun-ying, Tan Qian-cheng, et al. Multi-attention fusion network for medical image segmentation [J]. Journal of Computer Applications, 2022, 42(12): 3891-3899. | |
[17] | Wang Z Y, Zhu H, Liu F. SMSTracker: a self-calibration multi-head self-attention transformer for visual object tracking[J]. Computers, Materials & Continua, 2024, 80(1): 605-623. |
[18] | Vasanthi P, Mohan L. A reliable anchor regenerative-based transformer model for x-small and dense objects recognition[J]. Neural Networks, 2023, 165: 809-829. |
[19] | Battiti R. First- and second-order methods for learning: between steepest descent and Newton's method[J]. Neural Computation, 2014, 4(2): 141-166. |
[20] | Sutskever I, Martens J, Dahl G, et al. On the importance of initialization and momentum in deep learning[C]// International Conference on Machine Learning (ICML). Atlanta, 2013: 1139-1147. |
[21] | Qian N. On the momentum term in gradient descent learning algorithms[J]. Neural Networks, 1999, 12(1): 145-151. |
[22] | Duchi J, Hazan E, Singer Y. Adaptive subgradient methods for online learning and stochastic optimization[J]. Journal of Machine Learning Research, 2011, 12: 2121-2159. |
[23] | Mohamed R, Amany M S. A modified Adam algorithm for deep neural network optimization[J]. Neural Computing and Applications, 2023, 35(23): 17095-17112. |
[24] | Sarker I H. Deep learning: a comprehensive overview on techniques, taxonomy, applications and research directions[J]. SN Computer Science, 2021, 2(6): 420-420. |
[25] | Dai J F, Qi H Z, Xiong Y W, et al. Deformable convolutional networks[C]// Proceedings of the IEEE International Conference on Computer Vision (ICCV). Venice, 2017: 764-773. |
[26] | 欧阳继红,王梓明,刘思光.改进多尺度特征的YOLO_v4目标检测方法[J]. 吉林大学学报(理学版), 2022, 60(6): 1349-1355. |
Ouyang Ji-hong, Wang Zi-ming, Liu Si-guang. Improved multi-scale feature method for YOLO_v4 target detection[J]. Journal of Jilin University (Science Edition), 2022, 60(6): 1349-1355. | |
[27] | 魏东飞,熊峰,孔维畅.改进YOLOv4的轻量化目标检测方法[J]. 计量与测试技术, 2022, 49(11): 18-22. |
Wei Dong-fei, Xiong Feng, Kong Wei-chang. Improved lightweight target detection method of YOLOv4 [J]. Metrology & Measurement Technique, 2022, 49(11): 18-22. |
[1] | 郝博, 徐新岩, 赵玉欣, 闫俊伟. 基于改进YOLOv8的铆接孔表面缺陷检测[J]. 东北大学学报(自然科学版), 2024, 45(11): 1595-1603. |
[2] | 成怡, 王阳. 基于SPWVD-STFT的海面弱目标检测方法[J]. 东北大学学报(自然科学版), 2024, 45(10): 1401-1408. |
[3] | 包妮沙, 韩子松, 于嘉欣, 韦丽红. 基于改进Faster-RCNN的露天煤矿开采区遥感识别方法[J]. 东北大学学报(自然科学版), 2023, 44(12): 1759-1768. |
[4] | 李娟莉, 魏代良, 李博, 文小. 基于深度学习轻量化的改进SSD煤矸快速分选模型[J]. 东北大学学报(自然科学版), 2023, 44(10): 1474-1480. |
[5] | 顾德英, 罗聿伦, 李文超. 基于改进YOLOv5算法的复杂场景交通目标检测[J]. 东北大学学报(自然科学版), 2022, 43(8): 1073-1079. |
[6] | 于哲舟, 刘岩, 刘元宁. 基于YOLOV3改进的虹膜定位算法[J]. 东北大学学报(自然科学版), 2022, 43(4): 496-501. |
[7] | 杨爱萍, 宋尚阳, 程思萌. 轻量化自适应特征选择目标检测网络[J]. 东北大学学报(自然科学版), 2021, 42(9): 1238-1245. |
[8] | 宋欣, 李奇, 解婉君, 李宁. YOLOv3-ADS:一种基于YOLOv3的深度学习目标检测压缩模型[J]. 东北大学学报(自然科学版), 2021, 42(5): 609-615. |
[9] | 魏颖, 徐楚翘, 刁兆富, 李伯群. 基于生成对抗网络的多目标行人跟踪算法[J]. 东北大学学报(自然科学版), 2020, 41(12): 1673-1680. |
[10] | 王生生, 王萌, 王光耀. 基于深度神经网络剪枝的两阶段遥感图像目标检测[J]. 东北大学学报:自然科学版, 2019, 40(2): 174-179. |
[11] | 朴春赫, 潘怡霖, 赵海, 朱宏博. 基于改进ViBe的多行人检测方法[J]. 东北大学学报:自然科学版, 2016, 37(4): 481-485. |
[12] | 徐久强, 江萍萍, 朱宏博, 左伟. 面向运动目标检测的ViBe算法改进[J]. 东北大学学报:自然科学版, 2015, 36(9): 1227-1231. |
[13] | 佟国峰,蒋昭炎,谷久宏,庞晓磊. 基于随机蕨丛的长期目标跟踪算法[J]. 东北大学学报:自然科学版, 2013, 34(1): 4-8. |
[14] | 于晓升;吴成东;陈东岳;田子恒;. 基于多特征融合的遥感图像河流目标检测算法[J]. 东北大学学报(自然科学版), 2012, 33(11): 1547-1550. |
[15] | 闫爱云;李海朋;李晶皎;王骄;. 视频运动目标提取的实现[J]. 东北大学学报(自然科学版), 2011, 32(11): 1558-1561. |
阅读次数 | ||||||
全文 |
|
|||||
摘要 |
|
|||||