VEHICLE SAFETY SYSTEM : DEEP LEARNING BASE LICENSE PLATE RECOGNITION FROM VEHICLE AND MOTORCYCLE MIXED FLOW, AND LANE DEPARTURE EMBEDDED SYSTEM IMPLEMENTATION

• Main Authors: Wu, Yu-Hung, 吳宇鴻
• Other Authors: Hsieh, Jun-Wei
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/88hhha

碩士 === 國立臺灣海洋大學 === 資訊工程學系 === 106 === Two novel vehicle-related systems are proposed in this thesis; that is, deep learning based-on license plate recognition system and lane departure warning (LDW) system. For the first one, it can recognize license plates not only from cars but also motorcycles. There are two special cases we cannot recognize the letters of plates. One is the system cannot identify these license plates immediately. Another is the skewed plate cause missing characters or unrecognizable in the system for motorcycle’s license plate. In order to solve these problems, we proposed the deep learning technology to identify each vehicle’s license plate, the approach can re-organize and discriminate each founded character in the license plate. In traditional method only locate the license plate and cannot recognize it belong to car or motorcycle. We use the deep learning method to identify the license plate, such as the seven-code for the car or the six-code of motorcycle and find out the optimal result of character re-arrangement according to the regulation on license plate. For the second LDW system, we use the innovative CPU+IVE hardware intelligent algorithm with accelerated engine technology by Hisilicon Semiconductor to improve the problem of low frame rate. Filter out non-lane lines by calculating the slope, the length and the position of each line to, the most appropriate left and right lane lines could be screened and selected. When one of the lane lines disappeared or continued to deviate, a warning alarm would be sounded to remind the driver.