Robotic navigation and obstacle avoidance with computer vision and deep learning

With recent advancements in technology, deep learning is now able to be applied in many areas. With Convolutional Neural Networks, information from images can be extracted, which can then be used to learn more about the surrounding environment. In this project, we attempt to replicate the success of...

Full description

Saved in:
Bibliographic Details
Main Author: Heng, Shaun Wei Quan
Other Authors: Xie Lihua
Format: Final Year Project
Language:English
Published: 2019
Subjects:
Online Access:http://hdl.handle.net/10356/78049
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Nanyang Technological University
Language: English
Description
Summary:With recent advancements in technology, deep learning is now able to be applied in many areas. With Convolutional Neural Networks, information from images can be extracted, which can then be used to learn more about the surrounding environment. In this project, we attempt to replicate the success of the NVIDIA Redtail Team, where a method to autonomously fly a UAV for over a kilometre with a low-cost web camera and the Jetson TX1 was presented. However, with the lack of forest trails in NTU, the corridors of NTU will be used as the environment instead. A self-trained image classification CNN and a pre-trained object detection CNN are implemented on a UGV with self-written ROS nodes, and the report evaluates the method presented by them as well as the performance of some of the traditional computer vision algorithms used by us to obtain more information about the surrounding of the robot. To train the image classification CNN, data was gathered. The procedure to gather the data as well as the training process will be described in the report.