Comparative Study on Obstacle Detection and Avoidance System by using Real-Time Image Processing and Artificial Intelligence in Autonomous Wheelchair Application

Tan, Shang Deen (2018) Comparative Study on Obstacle Detection and Avoidance System by using Real-Time Image Processing and Artificial Intelligence in Autonomous Wheelchair Application. Final Year Project (Bachelor), Tunku Abdul Rahman University College.

Text TAN SHANG DEEN - FULL THESIS.pdf Restricted to Registered users only Download (3MB)

Abstract

The main research in this project is to develop an Artificial Intelligence based obstacle detection system in relation to obstacle avoidance system of an autonomous wheelchair, while at the same time to produce a comparative study between Real-Time Image Processing based obstacle detection algorithm and said Artificial Intelligence algorithm. The proposed obstacle detection system is achieved through the application of camera sensor with the implementation of Artificial Intelligence techniques in image processing. Convolutional Neural Network approach is chosen for the implementation of Artificial Intelligence (object recognition) to the obstacle detection system. A pre-trained Convolutional Neural Network model known as MobileNet SSD, and deep neural network (dnn) module in OpenCV library (for live video streams) are utilized in developing the object recognition algorithm. A pair of DC brushed motors and a pair of Smart Drive 40 motor drivers made up the motion control system that acts correspondingly to the inputs obtained from the obstacle detection system(s). Comparative researches on the performance of Real-Time Image Processing based obstacle detection algorithm and proposed Artificial Intelligence based obstacle detection algorithm are carried out through experiments and analysis on the relevant Performance Parameters and Performance Measures such as Light Intensity, Static Obstacle, Dynamic Obstacle, Wall Detection and algorithm respond time. The results of the experiments are analyzed and discussed. The results obtained indicate that while the performance of the Artificial Intelligence based obstacle detection algorithm is not particularly sensitive to the light intensity, it has a relatively low performance in terms of reaction speed and wall detection when compared to the Image Processing based obstacle detection system.