Automatic mapping of traffic signs

Nowadays, people easily can get into their cars and drive hundreds of kilometers in a few hours, but for that to work efficiently a system of rules must be applied and those rules have to be communicated transparently. This is why traffic signs are an influential part of our lives and every kind of information about each is helping the government, the community, and the drivers. This paper presents a novel and cost-efficient method for acquiring information on traffic signs, such like the category and the 3D position. The former can be gained using camera images and a Convolutional Neural Network model. The latter can be obtained using positioning devices. With the help of a GNSS device the absolute position of the vehicle can be learned and based on that a local coordinate system can be established. From the vehicle's point of view the coordinates and the orientation of the traffic sign can be acquired by applying a stereo camera and an IMU (Inertial Measurement Unit) sensor. Then, with the help of these attributes a large database can be built, maintained, and updated. This project displays that adequately precise data can easily be accessible using a few cheap devices and sensors.