Publication: A prototypic implementation of self-driving cars based on computer vision and deep learning
Date
2019
Authors
Nurul Fatin Nadiah Abdul Aziz
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Malaysia is one of the developed countries in South Asia. Proves that the local technologies in automobile industry growth widely such as Proton Holding and Produa Sendirian Berhad became one of the factors on selecting the title, a prototypic self-driving car using deep learning and computer vision. Besides that, the idea of this project also coming from many traffic congestions and accidents throughout the year recorded all over the world including Malaysia. Thus, to overcome this issue, this research project is purposed. This project focusing on three main goals which to train a set of data using neural network and achieve 70% of accuracy, design a self-driving car using remote-control (RC) car, Raspberry Pi, Pi Camera and Arduino and implementing a deep learning and computer vision and make it an autonomous RC car. Deep learning architecture was introduced to train set of data and achieve up to 70% accuracy. In this project, two models were presented. The models are You Only Look Once (YOLO) and Visual Geometry Group 16 (VGG16), the sub architecture from Single Shoot Multibox Detector. YOLO model is a model mean for training from scratch technique while VGG16 model is a model used for fine-tuning. As a result, VGG16 model is producing high accuracy compare to YOLO model. In designing a self-driving car using remote-control (RC) car, few components are involved. The components used are Raspberry Pi, Arduino, Pi Camera, RC car and Ultrasonic sensor. The connection of the elements is discussed in subsequence chapter under Methodology part. This experiment was differentiating between two design where the first design was constructed using commercial RC car available in the market and the second design was built by assembling mentioned components. The last objective is implementing a deep learning and computer vision and make it an autonomous RC car by simply combining the trained dataset with the designed RC car.
Description
TL152.8.N87 2019
Keywords
Autonomous vehicles , Computer vision