The speed limit signs recognition algorithm system installed in mobile devices has real-time performance, which can warn drivers of speed limit and improve traffic safety. Therefore, this paper proposes a real-time recognition algorithm for speed limit signs based on convolution neural networks and the system development combined with vehicle tests is carried. Firstly, the color segmentation is carried out based on HSV spatial color model, and the images are preprocessed by the morphological filtering. Then in order to locate accurately the position of speed limit signs , the speed limit signs’ gray scale and digital distribution characteristics are used to exclude the pseudo-targets. Secondly, image enhancement and normalization are performed on the captured image to improve the clarity and tidiness of it. Thirdly, the convolution neural networks model of LeNet-5 is constructed which can identify the speed limit signs. Fourthly, The system of embedding the convolutional neural networks model in Caffe deep learning framework is developed, and the GTSRB image data set is trained and tested through it. The iteration results show that the test accuracy rate reach 97%.Finally,the system development based on Android platform combined with vehicle tests is carried out. The results show that the accuracy rate of the real-time recognition and warning system for speed limit signs installed in mobile devices is more than 90% at different speeds and the maxium accuracy rate can reach 94%.Further,through the test process, found that it has strong real-time performance, system stability and recognition robustness. Keywords: speed limit signs, real-time recognition algorithm, convolutional neural networks, traffic safety, warning

CICTP