LiDAR Point Cloud Object Tracking

Overview

This project focuses on tracking objects in LiDAR point cloud data using boundary boxes. The tracking is performed by measuring distances between frames, predicting future positions with a Kalman filter, and comparing velocities to improve tracking accuracy. Additionally, it provides a visualization tool to generate a bird's eye view (BEV) of the tracked objects as a video or GIF.

Input Format

The input to the tracking system is a list of boundary boxes, each defined by 7 parameters:

• 3 values for the center of the box (x, y, z)
• 3 values for the extent of the box (x, y, z)
• 1 value for the yaw angle (rotation)

Output

1. Object Tracking: The tracking system outputs a 3d_ann.json file that contains the boundary boxes and their associated tracking IDs.
2. Bird's Eye View (BEV) Visualization: This tool generates a BEV video or GIF of the object tracking results using the LiDAR data.

Key Features

1. Object Tracking:

File name: object_tracking.py

• 2D Space Tracking: Tracks objects based on the closest distance in the X, Y 2D space. The z (height) is ignored for tracking.
• Kalman Filter: A Kalman filter is used to predict the next position of a boundary box, enhancing tracking accuracy.
• Velocity Comparison: Velocity comparison between consecutive frames is used to improve the object association and reduce ID switching errors.

2. Bird's Eye View Visualization:

File name: cadc_devkit/run_demo_lidar_dev2.py

• BEV Video/GIF: Generates a birds-eye view representation of the object tracking results, allowing for visual inspection.

Installation

Install the required dependencies by running the following commands:

pip install -r requirements.txt
pip install json
pip install re

Sample Out: