A ROS package for tracking people in RGB-D images with SORT (Simple Online and Realtime Tracking) algorithm.
To obtain a bachelor's degree through a stay at CIMAT, this project was carried out. Under the supervision of a CIMAT researcher.
Developing improvements for the project ...
$ cd ~/catkin_ws/src
$ git clone https://github.com/JCLArriaga5/cimat_sort_rgbd.git
$ cd ..
$ catkin_make
The program pcl_people_detector is the code created by PCL to detect people on a ground plane with RGB-D data, to test it requires having connected an rgbd sensor (e.g. Kinect, Realsense).
Execute following command:
$ rosrun cimat_sort_rgbd pcl_people_detection
To simulate the plc detector, using an RGB-D image dataset it is necessary to download the dataset previously. For the simulation with a dataset, we make use of the EPFL-LAB dataset, here you can see where to download it.
Once the dataset is downloaded in the corresponding folder, you can execute the following command to run the simulation with the dataset
$ rosrun cimat_sort_rgbd dataset_pcl_people_detection
The state vector of each objective for the Kalman filter that was used was the following:
where are the coordinates of the centroid of the enclosing box or cylinder, taken from the top or bottom. The height the radius were taken as constants, so are the respective velocities of the centroid. For the data association, an IoU of the upper circles of the cylinder of each objective was used.
To test the SORT algorithm with EPFL-LAB dataset and PLC detector use the following command.
$ rosrun cimat_sort_rgbd dataset_sort_test