Real time road traffic sign detection and recognition systems using Convolution Neural Network on a GPU platform
Lai Jia Shyan, Tiong Hoo Lim, Dk Norhafizah Pg Hj Muhammad
Abstract
Artificial Intelligence has been widely used for object detection in various applications such as text recognition and wildlife detection. Although it can detect an object at a certain level of confidence, the accuracy can be affected by the object's background, lighting condition, or when the object is moving. In this paper, a machine learning algorithm is applied to develop a traffic sign detection and recognition system (TSDR) using Convolution Neural Network (CNN) to build a localised traffic sign model using a novel multi-objects suppression approaches. The constructed model is tailored for specific country's traffic signs and can be installed on a portable microcontroller with GPU capability, to provide a real-time traffic sign recognition system. The proposed system can update the traffic sign model online or offline. Extensive experimental evaluations have shown that the traffic sign's model can be customised using multi-objects suppression with CNN and the accuracy of the model increases with the number of training samples.