Recognition of traffic marks in the images of WAHD lens by using color information and neural networks

In their laboratory, the authors have conducted research into a special super wide angle with high distortion lens (WAHD lens) which is designed to be functionally similar to the human eye. By using this lens, they optically obtain foveated information (distorted image). Color information and neural networks are used to make a computer recognize the traffic marks from the distorted image. This paper describes a color characteristic compensation method for the image obtained by WAHD lens, and a feature generation method based on discrete cosine transformation (DCT). The features are used in backpropagation trained neural networks. They conclude that this approach can be used in robots provided with wide angle vision sensors with high distortion lens to recognize traffic markings effectively.