https://github.com/shaoshengsong/DeepSORT
support deepsort and bytetrack MOT(Multi-object tracking) using yolov5 with C++
https://github.com/shaoshengsong/DeepSORT
deepsort track
Last synced: 3 months ago
JSON representation
support deepsort and bytetrack MOT(Multi-object tracking) using yolov5 with C++
- Host: GitHub
- URL: https://github.com/shaoshengsong/DeepSORT
- Owner: shaoshengsong
- Created: 2019-05-09T06:26:54.000Z (about 6 years ago)
- Default Branch: master
- Last Pushed: 2023-07-14T16:48:01.000Z (almost 2 years ago)
- Last Synced: 2024-10-27T20:21:09.278Z (8 months ago)
- Topics: deepsort, track
- Language: C++
- Homepage:
- Size: 79.1 KB
- Stars: 824
- Watchers: 7
- Forks: 192
- Open Issues: 46
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# DeepSORT
# MOT(Multi-object tracking) using yolov5 with C++ support deepsort and bytetrack
flyfish
## 前言
代码采用C++实现，目标检测支持YOLOv5 6.2,跟踪支持deepsort and bytetrack。
检测模型可以直接从YOLOv5官网，导出onnx使用
特征提取可以自己训练，导出onnx使用，onnxruntime cpu 推理，方便使用.
特征支持自定义维度例如 128,256,512等
本文源码地址
```c
https://github.com/shaoshengsong/DeepSORT
```
## deepsort v1.12
新增bytetrack跟踪
bytetrack论文
```c
http://arxiv.org/abs/2110.06864
```
bytetrack代码
```c
https://github.com/ifzhang/ByteTrack
```
## deepsort v1.1
deepsort原论文地址
```c
https://arxiv.org/pdf/1703.07402.pdf
```
```c
MOT using deepsort yolo5 with C++
```
操作系统：Ubuntu 18.04
### 版本更新说明
去除了TensorFlow依赖
为了不依赖硬件GPU，无需cuda，cudnn，更容易编译，使用PC版本。
为了更方便编译，采用CMakeList.txt。
### 依赖的库
opencv，可以下载opencv-4.6编译安装
Eigen3安装
```c
sudo apt-get install libeigen3-dev
```
onnxruntime，可以直接解压使用，无需编译
目标检测模型下载地址
```c
https://github.com/ultralytics/yolov5
```
网盘中有已经导出完成的模型
### 文件下载
百度网盘
链接：`https://pan.baidu.com/s/1igjNK2ty-H5AU_Ut08pkoA`
提取码：0000
内容包括
```c
cmake-3.21.4-linux-x86_64.tar.gz
onnxruntime-linux-x64-1.12.1.tgz
coco_80_labels_list.txt
opencv-4.6.0.zip
DeepSORT
yolov5s.onnx
feature.onnx
yolov5x.onnx
```
### 使用方法
#### 1 onnxruntime
设置自己的onnxruntime的解压目录
```
set(ONNXRUNTIME_DIR "/home/a/lib/onnxruntime-linux-x64-1.12.1")
```
#### 2 模型配置
以下三项根据自己的需要更改
文件`tracker/deepsort/include/dataType.h`
```c
const int k_feature_dim=512;//feature dim
const std::string k_feature_model_path ="./feature.onnx";
const std::string k_detect_model_path ="./yolov5s.onnx";
```
#### 3 主函数
选择打开视频文件或者视频流等
```c
cv::VideoCapture capture("./1.mp4");
```
### 扩展方式
1 整体分为两部分，新增检测模块放置detector文件夹，新增跟踪模块放置tracker文件夹
## deepsort v1.0
### MOT using deepsort yolo3 with C++
操作系统：Ubuntu 18.04
编译环境：Qt 5.12.2
深度学习的模型分两块，一个是目标检测，另一个是目标跟踪
#### 目标检测的模型
地址：`https://pjreddie.com/darknet/yolo/`
#### 目标跟踪模型
mars-small128
OpenCV DNN加载YOLO模型，不依赖Darknet库，cuda，cudnn
依赖Tensorflow，目标跟踪的特征部分使用TensorFlow C++的api。
OpenCV的安装可以参考
地址: `https://blog.csdn.net/flyfish1986/article/details/89157368`
Tensorflow的安装可以参考
地址：`https://blog.csdn.net/flyfish1986/article/details/89406211`
[多目标跟踪论文 Deep SORT 解读](https://flyfish.blog.csdn.net/article/details/89852370)
[多目标跟踪论文 Deep SORT 实现](https://flyfish.blog.csdn.net/article/details/90034289)
[多目标跟踪论文 Deep SORT 数据集说明](https://flyfish.blog.csdn.net/article/details/90070639)
[多目标跟踪论文 Deep SORT 特征提取CNN Architecture](https://flyfish.blog.csdn.net/article/details/90642532)
[多目标跟踪论文 Deep SORT 特征训练PyTorch实现](https://flyfish.blog.csdn.net/article/details/90702620)
[多目标跟踪论文 Deep SORT 特征训练TensorFlow实现](https://flyfish.blog.csdn.net/article/details/90379444)
[多目标跟踪论文 Deep SORT 评测指标](https://flyfish.blog.csdn.net/article/details/90200171)
[匈牙利算法](https://flyfish.blog.csdn.net/article/details/104298521)
[卡尔曼滤波 - 方程组转换为矩阵形式](https://flyfish.blog.csdn.net/article/details/118635703)
[卡尔曼滤波 - 一个方程背后的样子](https://flyfish.blog.csdn.net/article/details/118636055)
[卡尔曼滤波 - 匀变速直线运动](https://flyfish.blog.csdn.net/article/details/118613382)
[卡尔曼滤波 - 冥冥之中自有定数的正态分布](https://flyfish.blog.csdn.net/article/details/116067569)
[卡尔曼滤波 - 数据融合 data fusion](https://flyfish.blog.csdn.net/article/details/118613307)
[卡尔曼滤波 - 当前均值与上一次均值的关系](https://flyfish.blog.csdn.net/article/details/117931292)
[卡尔曼滤波 - 状态空间模型](https://flyfish.blog.csdn.net/article/details/118636364)
[卡尔曼滤波 - 5个公式出现的顺序](https://flyfish.blog.csdn.net/article/details/118709808)