3D Multi-Object Tracking using Lidar

Introduction to Lidar, Autonomous Driving Public Dataset Types, State of the Art Object Recognition Models for Lidar, and a Kalman Filter-based Tracking Algorithm is explained in this tutorial

Yağmur Çiğdem Aktaş

--

Hello everyone! I have completed my master's program after a very busy last 6-months period where I was having my internship and preparing my master's thesis report. Without surprise, my project was about “3D Multi-Object Tracking using Lidar for Autonomous Driving” and as a very beginner in Lidar, Tracking, and Autonomous Driving subjects, I decided to share my notes before I forget the details that I had trouble understanding as a beginner :)

This post will have the following structure to explain my project from top to bottom :

  • Background Information
    1. RGB-D Camera
    2. Point Clouds
    3. Object Tracking Types
  • Object Detection using Lidar
    1. PointNet
    2. VoxelNet
    3. SECOND
    4. PointPillar
    5. Model Selection
  • Kalman Filter-based Tracking Algorithm
    1. Main Types of Tracking Methods
    2. Kalman Filter
    3. Mahalanobis Distance
    4. Greedy Match Algorithm
    5. Optimizations
    6. Results

The main GitHub repo used for this project is public and accessible 😋

The background information part starts just below 🔽

The multi-object tracking (MOT) system performs accurate tracking of obstacles moving in front of or in the surrounding environment of an autonomous vehicle, including vehicle path tracking, non-motor vehicle trajectory tracking, pedestrian trajectory tracking, etc. This subsystem helps self-driving cars make decisions and avoid collisions with objects that may move. The main task of the multi-object tracking algorithm is to track many objects simultaneously and assign and maintain a corresponding ID for each object, which…

--

--