3D Object Detection and Velocity Estimation for Autonomous Vehicles

The idea is to not use lidar for depth and velocity estimation for AVs. To achieve this, implemented a model which combines 3D object detection and constant velocity Kalman filter tracker for AVs.

Project Website

Topic Covered:

  1. Basics of 3D object detection, constant velocity Kalman filter and Hungarian Algorithm.
  2. Backbone - CenterNet.
  3. Optimization and loss functions - Inverse sigmoid, Uncertainty based L1 loss, Behru loss, Scale Invariant loss.

Programming Language: Python, Pytorch