Based on Deep Learning for Forward Collision Warning and Lane-Mark Recognition Systems

• Main Authors: ZHOU, YUE-HAN, 周岳瀚
• Other Authors: PAI, NENG-SHENG
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/5wz2m8

碩士 === 國立勤益科技大學 === 電機工程系 === 107 === This thesis aims to develop a driver assistance system that provides forward collision warning and lane-mark recognition systems; the required network model of the system is achieved using deep learning (DL) technology. First, a webcam is used to capture the image ahead of the driver. Second, the image data are transmitted to the processing server for target recognition. Finally, the system data are displayed on the smart glass through the internet to achieve the augmented reality (AR) effects. The structure used herein is based on Tiny YOLO in YOLOv2 with the aim to reduce network complexity and improve computing efficiency. Additionally, K-means is used to select the anchor boxes for each dataset during the training processes. By assigning basis to the size of the forecast frame, the computational proficiency is improved. Finally, through the implementation of this system, the driver can instantly see the road conditions and lane-mark ahead. Thus, this system will reduce vehicle collisions and traffic violations, thereby ensuring driving safety.