Self-Driving System Implementation with Deep Learning and Image Classification

• Main Authors: Yow-Lin Wu, 吳侑霖
• Other Authors: Po-Chiang Lin
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/9ucbtd

碩士 === 元智大學 === 通訊工程學系 === 106 === Self-Driving system is constantly being researched, and it is a trend in the future. Now, there is an unmanned vehicle and automatic driving assistance system,and it has brought many convenience to humans. In this paper, we proposed a Self-Driving system that combines deep learning with image classification. Through the integration of software and hardware, it can be run in the real environment. The architecture includes deep learning, sensor integration, machine vision, and control system. Convolution Neural network is training image classification model and scene classification model, using lidar as obstacle avoidance, webcam as a images receiver, Arduino as direction controller, RC car as car body, and finally NVIDIA Jetson TX2 (embedded development board) as the core of operation. In the corridors of the building, this system can be automatically self-driven. This system is verified that the training of the neural network is very helpful for Self-Driving system. It also found that it have a good effect even if the number of training images do not reach tens of thousands of images. The complete implementation of the physical architecture proves that this system can work.