Final competition - Tracking

This task is based on tracking by detection method. We already have detection results (Bounding boxes) from detector. Our goal is to develope a good tracking algorithm.

Data preparation

• Detection result and frames meta data from google drive.
• If you want to evaluate and visualize, you will need to download nuscenes trainval dataset from Nuscenes website.
• The folder structure should be organized as follows before processing.

nusc_tracking
├── configs
├── docker
├── ros_ws
├── tools
├── data
│   ├── nuscenes
│   │   ├── maps
│   │   ├── samples
│   │   ├── sweeps
│   │   ├── v1.0-trainval

Environment setup

First of all, cd to your workspace and,

git clone https://github.com/derekray311511/nusc_tracking.git  

cd nusc_tracking 
# Build the docker imagecd nusc_tracking 
cd docker && docker build . -t tracking
# Run the docker image and create a container
bash run.sh

If you want to run the same container in other terminal:

docker exec -it tracking bash

Create a virtual path to data folder in docker

cd /home/Student/Tracking
ln -fsv /data data

Put the detection_result.josn and frames_meta.json (download from PART1) into your data folder

Tracking and Evaluation

# Arguments can be set in track_template.sh
bash tools/track_template.sh

Visualization

Need to download dataset from nuscenes trainval dataset from Nuscenes website. We use ros to visualize our tracking results.

cd ros_ws
catkin_make -DPYTHON_EXECUTABLE=/usr/bin/python3
source devel/setup.bash
# First terminal
roscore
# Second terminal
rviz -d configs/track.rviz # can run out of the docker container
# Third terminal
bash src/visualize.sh