没有合适的资源?快使用搜索试试~ 我知道了~
基于YOLOv5和DeepSORT的危险车辆目标轨迹追踪
需积分: 0 3 下载量 162 浏览量
2023-07-26
19:37:39
上传
评论
收藏 3.74MB PDF 举报
温馨提示
试读
68页
基于YOLOv5和DeepSORT的危险车辆目标轨迹追踪
资源推荐
资源详情
资源评论
说明书
目录
摘要..................................................................................................................................... I
Abstract
...............................................................................................................................II
1 绪论 ................................................................................................................................. 1
1.1 研究背景与意义 .................................................................................................. 1
1.2 国内外研究现状................................................................................................. 2
1.2.1 相关目标检测算法研究 ........................................................................... 2
1.2.1.1 传统目标检测算法....................................................................... 2
1.2.1.2 基于深度学习的目标检测算法 ....................................................3
1.2.2 相关目标追踪算法的研究 ....................................................................... 6
1.3 本文主要研究内容 .............................................................................................. 7
2
目标检测算法的相关理论............................................................................................. 9
2.1 卷积神经网络..................................................................................................... 9
2.1.1
卷积层...................................................................................................... 9
2.1.2 池化层 ....................................................................................................11
2.1.3 激活函数层 ............................................................................................. 12
2.1.4 损失函数................................................................................................ 13
2.2
基于候选区域的经典目标检测算法............................................................... 14
2.2.1 R-CNN .....................................................................................................15
2.2.2 SPP-Net
................................................................................................... 15
2.2.3 Fast R-CNN ............................................................................................. 16
2.2.4 Faster R-CNN
..........................................................................................17
2.3
基于回归的经典目标检测算法....................................................................... 18
2.3.1 YOLO v1 ................................................................................................. 18
2.3.2 SSD ..........................................................................................................19
2.3.3 YOLO v2
................................................................................................. 19
2.3.4 YOLO v3
和
YOLO v4
........................................................................... 20
2.4
本章小结........................................................................................................... 20
3
基于
YOLO v5
的目标检测算法 ................................................................................ 21
3.1
图像预处理........................................................................................................ 21
3.1.1 图像的几何变换 ..................................................................................... 21
3.1.2 mosaic(四图拼接) ................................................................................... 23
3.1.3 Mixup
数据增广 ..................................................................................... 23
3.1.4 HSV
空间变换........................................................................................ 24
3.2 YOLO v5
网络结构 ........................................................................................... 25
说明书
3.2.1 融合卷积 Conv ...................................................................................... 25
3.2.2 Focus 模块 ...............................................................................................26
3.2.3 SPP 模块 ................................................................................................. 27
3.2.4 C3 模块 ................................................................................................... 27
3.3 预测输出原理................................................................................................... 30
3.3.1 K-means
聚类算法..................................................................................30
3.3.2
正样本扩充............................................................................................ 31
3.3.3
目标框回归原理.................................................................................... 32
3.3.4 NMS
非极大值抑制 ............................................................................... 32
3.4
损失函数........................................................................................................... 33
3.5 本章小结........................................................................................................... 35
4 基于 DeepSORT 的多目标追踪算法 ...........................................................................36
4.1 多目标追踪算法 SORT.................................................................................... 36
4.1.1 卡尔曼滤波 ............................................................................................. 36
4.1.2 匈牙利算法 ............................................................................................. 37
4.1.3 基于 SORT 算法的多目标追踪工作流程 ............................................ 38
4.2
基于
SORT
的改进算法
DeepSORT
................................................................39
4.2.1 数据关联匹配策略................................................................................ 40
4.2.1.1 运动模型的度量 ..........................................................................40
4.2.1.2 外观模型的度量 ..........................................................................40
4.2.1.3 综合度量匹配 .............................................................................41
4.2.2 长时遮挡匹配策略 ................................................................................. 42
4.3 基于 DeepSORT 算法的多目标追踪工作流程 ................................................42
4.4 本章小结........................................................................................................... 43
5 实验过程与结果分析 ................................................................................................... 44
5.1 实验环境介绍................................................................................................... 44
5.2 目标检测评价指标以及实验参数设定........................................................... 45
5.2.1 目标检测评价指标介绍 ........................................................................ 45
5.2.2 实验参数设定 ........................................................................................46
5.3 实验过程与结果分析 ........................................................................................ 47
5.3.1 危险车辆数据集的建立 ........................................................................ 47
5.3.2 危险车辆数据集目标分析 .................................................................... 48
5.3.3 训练结果分析 ........................................................................................49
5.4 实际道路场景下危险车辆的检测结果分析与改进....................................... 52
5.4.1 检测结果与评价 .................................................................................... 52
5.4.2 改进后的检测结果与评价 .................................................................... 53
说明书
5.5 目标追踪实验结果与评价............................................................................... 56
6 结束语.......................................................................................................................... 58
参考文献
...........................................................................................................................
59
致谢
...................................................................................................................................
61
附录
...................................................................................................................................
62
说明书
I
摘要
目标检测与追踪是计算机视觉领域重点研究的课题,也是现今社会各领域倍受关注
的话题。基于当下目标检测技术已经近乎成熟,但是在车辆目标检测方面仍旧是大范围,
多类别的检测与分类任务,存在检测针对性不强以及目标不明确等弊端,有着极大的改
进空间。本课题是完成基于 YOLO v5 和 DeepSORT 进行危险车辆目标追踪的任务,通过
利用基于卷积神经网络的 YOLOv5 做检测器对危险车辆进行检测与分类,然后利用
DeepSORT 算法实现视频序列前后帧或者多帧间的数据关联从而完成对目标的追踪任
务。通过完成本课题任务,后期可以继续研究如何嵌入到车载系统的移动端,对生活中
车辆在行进过程中遇到一些存有交通安全隐患的危险车辆时能够起到提前预警驾驶员
的作用具有一定的现实借鉴意义。
关键词:卷积神经网络;YOLO v5 ;DeepSORT;数据关联;目标追踪
说明书
II
Abstract
Object detection and tracking is a key research topic in the field of computer vision, and
it is also a topic of great concern in various fields of today's society. Based on the current
target detection technology is almost mature, but it is still a large-scale, multi-category
detection and classification task in vehicle target detection.This topic is to complete the task
of tracking dangerous vehicle targets based on YOLO v5 and DeepSORT. By using YOLOv5
based on convolutional neural network as a detector to detect and classify dangerous
vehicles,Then, the DeepSORT algorithm is used to realize the data association between the
frames before and after the video sequence or between multiple frames, so as to complete the
tracking task of the target.By completing the tasks of this topic, we can continue to study how
to embed the mobile terminal of the vehicle system in the later stage,It has a certain practical
reference significance for the vehicle in life when it encounters some dangerous vehicles with
traffic safety hidden dangers in the process of traveling.
Key words:Convolutional Neural Network;YOLO v5;DeepSORT;Data Association;
Object Tracking
剩余67页未读,继续阅读
资源评论
@小菜鸡
- 粉丝: 0
- 资源: 1
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功