Traffic Warning Sign Recognition System Using Computer Vision and Machine Learning

The paper is devoted to the development and testing of a traffic warning sign recognition system. The implementation using cascade classifier as detector and convolutional neural network as recognizer was proposed. Hardware part of the system was implemented on the basis of single board computers. The influence of applied single board computer architectures and usage of neural hardware accelerators on the system parameters was analyzed. The proposed system can be used to detect and recognize traffic warning signs in real time with high performance and accuracy. The result detection accuracy is 83% with recall of 84% and recognition accuracy is 98.85%. The result time for one frame is 250 milliseconds with a total distance up to 15 meters.