Multi-camera-based obstacle-detection algorithm for robots

• Main Author: Han, Mingyu
• Other Authors: Wang Dan Wei
• Format: Thesis-Master by Coursework
• Language: English
• Published: Nanyang Technological University 2022
• Subjects: Engineering::Computer science and engineering::Computing methodologies::Image processing and computer vision
• Online Access: https://hdl.handle.net/10356/155132
• Institution: Nanyang Technological University

The main content of this paper is the process and results of algorithm research for robots based on multi-channel input data. Robots and automated driving are more and more popular, but some of the single-camera information is not enough to guide the regular work of the robots or automatic driving vehicles. At the same time, robots or automatic driving vehicles for obstacles, such as pedestrians object detection, need to improve, so this paper's research purpose is to increase the depth completion ability of the algorithm from the perspective of a deep learning algorithm. The method adopted in this paper is to modify the deep learning model based on previous studies and conduct comparative experiments on their effects to explore better deep learning methods to achieve deep completion. The main work of this dissertation is as follows: First, it explores the relationship between layers of a deep learning network and data features. Secondly, It tests the effect of the attention mechanism on the deep completion problem. The results obtained in this paper are as follows: First, through experiments, it is found that on a small data set, a deeper model may not obtain better results. Secondly, it was found that the experimental results become better with the attention mechanism. Third, experiments show that combining multiple good ideas does not necessarily lead to better results. Keyword: Deep learning, depth completion.