person_tracking_ros

A ros package that tracks a selected target person using YOLOv3 and DeepSORT

Published Topics and Provided Services

Published topics

Topic Name	Message Type	Purpose
/person_tracking/bbox_center	geometry_msgs.Point	Publishes the centerpoint of the target bounding box when a person is selected as target
/person_tracking/target_angle	std_msgs.Float32	Publishes the angle of the target bounding box when a person is selected as target. Uses the person_tracker/camera_fov param declared in person_tracker.launch to compute angle.
/person_tracking/target_present	std_msgs.Bool	Publishes True if selected target person is present in the camera view. Publishes False if person is not present.
/person_tracking/target_indice	std_msgs.Int32	Publishes the indice of the selected target person. If no person is selected, publishes 0.
/person_tracking/detection_indices	std_msgs.Int32MultiArray	Publishes an array of all detection indices when no person is selected as target. Publishes target indice if target is selected.
/person_tracking/deepsort_image/compressed	sensor_msgs.CompressedImage	Publishes a compressed image of drawn bounding box of detections and track processed by tracker

Services

Service Name	Input Type and Name	Output Type and Name	Purpose
/person_tracking/clear_target	std_msgs.Bool clear	std_msgs.Bool success	Untracks selected target. Returns true if succeeded. Returns false if no error encountered or no track started
/person_tracking/choose_target	std_msgs.Int32 target	std_msgs.Bool success	Choose indice for tracking. Returns true if indice present in camera and track started. Returns false if no detection available with given indices or track already started

Setup

Requirements

Python modules

• Python=3.6
• CUDA 10
• numpy
• scipy
• opencv-python
• sklearn
• torch >= 0.4
• torchvision >= 0.1
• pillow
• vizer
• edict
• pyyaml
• easydict

ROS

• ROS Melodic
• Any ROS Package for accessing webcam. I use video_stream_opencv.

Installation

Setup ROS Melodic and your catkin_ws if you have not done so.

Install the required python 3 modules using pip3 or a virtualenv. Do not use Anaconda. Git clone this repository into src of your catkin_ws. Follow

cd catkin_ws/src
git clone https://github.com/khayliang/person_tracking_ros.git
cd person_tracking_ros

Download the YOLOv3 weights

cd src/weights
wget https://pjreddie.com/media/files/yolov3.weights

Download the reidentification weights into the folder from this google drive.

Compile the nms module

cd ..
cd detector/YOLOv3/nms
sh build.sh
cd ../../../..

After installing your python modules, source the python intepreter.

cd bin

Open the file person_tracker.py with your preferred text editor

If you're using virtualenv, change the first line in the code accordingly:

#!/[path-to-virtualenv]/bin/python

If you directly installed to your python3:

#!/usr/bin/env python3

Once all that is done, make the package.

cd ../../..
catkin_make

Configuring launch file

Open the launch file in person_tracking_ros/launch/person_tracker.launch

In this line:

  <node name="person_tracker" pkg="person_tracking" type="person_tracker.py" args="--img_topic /camera/image_raw/compressed" /> 

Change /camera/image_raw/compressed to your custom image topic. Only compressed image is supported.

Running

Initialization

Start your roscore

roscore

Start your camera stream node. Here, I'm using video_stream_opencv.

roslaunch video_stream_opencv camera.launch

Ensure that your image topic is correctly set to your preferred tracker. Now source your workspace and run the tracker.

source /devel/setup.bash
roslaunch person_tracking person_tracker.launch

Tracker should now be running.

Viewing tracker image output

The tracker publishes the image output in compressed form to the topic:

/person_tracking/deepsort_image/compressed

To select view the tracker output, use rqt_image_view

rqt_image_view

And choose the stated topic above to view the image.

Choosing tracking target

The tracker uses a service to choose a target to track:

/person_tracking/choose_target

To choose a target, run the following commands.

rosservice call /person_tracking/choose_target [ENTER DETECTION INDICE HERE]

Service will return True if track is sucessfully started.

Clearing tracked target

The tracker uses a service to clear the track

/person_tracking/clear_target

To clear a target, run the following command.

rosservice call /person_tracking/clear_target true

Service will return True if track is successfully cleared.