To see all references under review, please visit our CiteULike page.
Traffic sign recognition has been an important topic in various research areas including, but not limited to, video surveillance, GIS information collection, and federal asset management.
While numerous computer-vision based approaches are developed toward traffic sign detection (See Road signs recognition - survey of the state of the art), the methods for traffic sign recognition, which is identifying the exact class of traffic sign, one of the popular approaches, has been heavily depending on the artificial neural network considering. This is partially due to the noisy environment on the real road in the view of image processing which leads many research to take advantage of the well-known robustness of ANN to its application domain.
In this context, Liu and Ran presented the stop sign recognition work. They detected the stop sign from the given image by using the color segmentation on HSV color space and after check the object width, aspect ratio and symmetrical level to filter out the stop sign region candidates. This extract ROI image after resizing to 30x30 pixels is directly put into the ANN configured 30x30 (900) input nodes, 1 hidden layer with 6 hidden neurons, and 2 output neurons indicating “stop sign” or “not”. Their training curve shows the typical back propagation curvature showing the convergence. Their over whole experiment is simple and not creative compared to other works.
Fang. et. al presented a method combined of various techniques. They first construct the database of the traffic signs on a selected initial size and the image size of a road sign matches its initial size, it gets detected. After, a Kalman filter is used to track a sign until it is sufficiently large to be recognized as a specific standard sign. They extracted the centers of specific color regions from the HSI color space and simultaneously, an edge-detection method is applied to acquire gradient values in specific color regions to construct an edge image. Both color and edge features are extracted by each neural network. And both features are integrated for the detection using a fuzzy approach. These steps are repeated while tracking the traffic sign regions in the video. In their approach, interestingly, circle, octagon, and triangle type shapes are only concerned. However, most informative traffic signs have rectangle shapes but in fact, most general objects in the real world also have also have such rectangle shapes. They showed several successful cases but the experiment results on the specified number of sets with false alarms or recognition ratios are not provided.
After, Fang et. al extended their works by presenting more sophisticated approach that use the uniquely designed neural network that is composed in two separated ANN: first for the traffic sign ROI extraction and the second is for the traffic sign recognition. They composed the ANN architecture to represent the two dimensional property of image in which the value of the pixel is affected by their neighbors. This is a familiar concept in computer vision specifically using Markov random field modeling. So their ANN architecture if reflecting this generic image brightness and color behavior in training the ANN network. However, their actual recognition is very limited to the big category of traffic sign like whether they are warning signs or not.