YOLOV5 + Deepsort Tracking with ROS Noetic and Pytorch

---

Prepare

1. Install ROS Noetic

2. Create a virtual environment with Python >= 3.8

    conda create -n yolov5_deepsort python=3.8
    conda activate yolov5_deepsort

2. Install pytorch and torchvision

    pip3 install torch torchvision torchaudio

you can follow the official website to install the correct version of pytorch.

cuda 12.2, cudnn 8.9.7.29 is tested in this branch. You can try other versions, but I am not sure if it will work.

3. Install Python3 Dependencies

    pip3 install rospkg catkin_pkg

4. Clone the repository

• Clone the repository to your catkin workspace

    mkdir -p ~/catkin_ws/src && cd ~/catkin_ws/src
    git clone https://github.com/ChiRanTou/Yolov5_Deepsort_pytorch_ROS.git
    cd ..
    catkin_make

• Ensure that all dependencies are met.

    cd ~/catkin_ws/src/Yolov5_Deepsort_pytorch_ROS
    pip3 install -r requirements.txt

• You can download YOLOv5 weight from the official website of yolov5 and place the downloaded .pt file under yolov5/weights/. I've already put the yolov5s.pt in the folder. You can other weight file if you like.

• You may also need to download the deepsort weight file from here and place ckpt.t7 file under deep_sort/deep/checkpoint/. I've also already put the ckpt.t7 in the folder. You can the file if you like.

---

Run

Before running the Project, you may notice that a ROS simulation enviornment is required. A robot with rgb camera is also needed to send the sensor_msgs/Image topic. So you have to get one first.

1. open the launch file and change the image_topic to the topic that your camera publish the image.

    <arg name="image_topic" default="/rgb/image_raw"/>

2. start your ROS simulation enviornment and make sure the camera is working.

3. launch the task you want to run.

    # for dectection only
    roslaunch yolov5_deepsort detector.launch

    # for tracking
    roslaunch yolov5_deepsort tracker.launch

4. open rviz if you didn't open it, and add the detected_objects_image/IMAGE or tracked_objects_image/IMAGE based on your task to the display panel. You can now see the result in the rviz.

---

Reference

• yolov5_deepsort_ros
• DeepSORT_YOLOv5_Pytorch
• yolov5
• deep_sort

Note: please follow the LICENCE of YOLOv5!