Semantic Video Analysis and Lane Detection via Convolutional Neural Network for a Self Driving System

• Main Authors: YANG,JHE-WEI, 楊哲維
• Other Authors: CHEN,JU-CHIH
• Format: Others
• Language: zh-TW
• Published: 2017
• Online Access: http://ndltd.ncl.edu.tw/handle/j84mp3

碩士 === 國立高雄應用科技大學 === 資訊工程系 === 105 === It’s possible that within the next ten or twenty years, self-driving car will become common, and our roads might even become safer. There are millions of people die from traffic accidents every year, and most of these accidents are caused by human-error. So it’s possible that well-made self-driving cars could decrease the number. Self-driving car can also help people who couldn’t drive, like visually-impaired or elderly. People can use these time working or sleeping when driving. Self-driving cars need GPS to kwon the current location and destination, and they have a bunch of sensors to constantly scan the area around them, looking for any hazards weather another car, a bicycle, or a human. Google self-driving car sensors include a camera to see, and radar and lasers to map out the area. Our method focuses on using camera images to recognize all the objects in the area. It’s hard to recognize road, tree, human, symbol sign. But, deep learning is great for complex patterns recognition. With enough training data, deep learning can handle this kind of problem. After recognition, the system could know the available driving area and unavailable area, and then compute the available driving area border by RANSAC (Random sample consensus). Finally, analyze these result to get the suggest driving path.