Recognition of maximal speed limit traffic signs for use in advanced ADAS algorithms

Advanced Driver Assistance Systems (ADAS) have been increasingly developing, specifically in the last decade. One of such ADAS is that intended for traffic signs recognition. This paper deals with the recognition of a specific subset of traffic signs, i.e. speed limit traffic signs. The complete solution is based on the usage of machine learning and finally implemented in the C programming language. After the optimization process, the final solution is implemented on the real ADAS board, to check its performance in a real operational environment. Due to the limited resources of the ADAS board itself, a simple Convolutional Neural Network (CNN) was created to recognize speed limit traffic signs. For CNN training a large database of 6891 training images is used. When testing the solution, 731 test images from the real traffic are used, as well as 123 real video sequences. The test results show that in certain situations the proposed solution is capable of achieving high performance in terms of precision, while in some cases additional improvements of the solution should be investigated. It is capable of processing 12 frames per second when operating with state-of-the-art automotive camera resolution, i.e. 1280x720 pixels.