Faculty Supervisor

Thesis Supervisor: Dr. Ghassem Tofighi Title: Professor, School of Applied Computing

Date of Defense

Winter 12-10-2020

Program Name

Honours Bachelor of Computer Science (Mobile Computing)

School

Applied Computing

Keywords

e-scooter, obstacle detection, single-camera obstacle detection, depth estimation

Department

Faculty of Applied Science & Technology (FAST)

Description

The YOLO3 return the detected object recognition is not consistent at all, you can get it back form the accuracy and the TN and the FN return count, you will know that the error is not very impressive. And we found out it has a problem cannot keep on detect the same object in a sequence of consecutive frames. Actually, the obstacle is existing, but some of the frame cannot show and broke the sequence of the detected boundary box.

Abstract

Recently, the e-scooter is more popular and trendy, therefore improvement of the safety feature of e-scooter, a moving vehicle that has a maximum speed about 40 km/hour is very important. Especially, Toronto has just approved these kind ofvehicles that can ride on the bike lanes under the by-law. In this research, an algorithm is proposed to use a single camera for detecting the obstacles using the latest deep learning-based models. This new approach omits the complex alignment and expensive equipment. The deep learning models are employed to judge whether the safe distance exists between the e-scooter and the obstacle. If it detects the obstacles, it reports the closest obstacle by comparing depth information of all obstacles and a warning will be issued based on the obstacle detection. If no obstacle is detected, a message informing about the safe riding situation is announced. The proposed algorithm has the accuracy about 70%. In the future work, by including additional advanced deep learning models for depth estimation and distance calculation, a better performance is expected.

Creative Commons License

Creative Commons Attribution 4.0 License
This work is licensed under a Creative Commons Attribution 4.0 License.

Share

COinS