Dynamic Environment Perception for Autonomous System: Real-Time Approach to Obstacle Detection

This repository contains the necessary information and algorithm collected for obstacle detection using a monocular camera.

Appearance Based Detection

Model Implemented

The state-of-the-art model implemented for obstacle detectiona are as follows:

• YOLOv8 (✔)
• Detectron2
• FastSAM

Depth Based Algorithm

Models used for relative depth estimation are as follows:

• MiDAS 3.1
• Depth Anything (✔)
• Depth Anything V2 (To be Implemented)

Hardware used

Two Hardwares were used each having different purposes.

For Training Models

• AMD Ryzen 7 6800H + NVIDIA GeForce 3060 GPU GDDR6 @ 6GB

For Inference and Testing Models

The models selected were tested in the selected hardware to check its utility in a Real-Time Environment

• Jetson Nano Developer Kit 4GB
  • Ubuntu 20 for Jetson Nano with preinstalled requirements
  • Wireless TP-link Nano Adapter