Identification in the Presence of Speed Bumps through Few-Shot Learning

Ng, Ernest Tze-Jie (2022) Identification in the Presence of Speed Bumps through Few-Shot Learning. Final Year Project (Bachelor), Tunku Abdul Rahman University College.

Text Ernest Ng Tze-Jie.pdf Restricted to Registered users only Download (3MB)

Abstract

Speed bump is one of the major components in traffic management system, function as to slow down the driving speed of the vehicles, improves the safety of the road users, reduce rate of accidents as well as regulating the traffic. As technology evolves and advances, autonomous driving technology has seen a tremendous growth in the past decade and to achieve a higher level of driving automation, real-time road condition data must be collected and analysed accurately to ensure the safety of the drivers and passengers, including the detection of speed bumps on the road. However, only scarce amount of work had been done on the detection of speed bumps and mostly involve in the use of hardware sensors in achieving higher detection accuracy. While the image processing approach, despite able to achieve high accuracy, it requires large amount of data and time for model training. The few-shot learning image processing approach using Siamese network is introduced and developed to identify the presence of marked speed bump to overcome the limitations addressed above. The model is able to achieve high test accuracy with a low loss using Siamese network with contrastive loss using a smaller dataset.