Duration: Feb 2020 - May 2020
Purpose: This project was conducted at the UAV Lab, Indian Institute of Science, Bangalore
Skills: Image processing, computer vision, machine learning, control systems
This project attempts to create a simple low cost obstacle avoidance system for avoiding large obstacles such as buildings or trees.
Approach
For this project, we had groups of interns working under our guidance to try different parallel approaches. One group worked on Machine Learning with a monocular camera to obtain a continuous depth estimation. Another group used monocular cameras to detect key-point sizes using SURF and ORB algorithms. A third group used stereo cameras to generate an estimated depth map. We tested using lasers and and ultrasonic sensors for passive obstacle detection.
Each of these provided limited results, with obstacle avoidance achieved at a maximum speed of 1 m/s when the object was 2m away.
Unfortunately, this project was cut short by the pandemic before we were able to get any concrete results.