Biological Metaheuristics for Road Sign Detection
The automatic detection and recognition of traffic signs in road scenes has received particular attention for the past two decades, especially in the context of driver assistance systems design. Aimed at performing in real time, the methods which have been developed until now generally combine a segmentation phase based on color or shape and a classification phase. Despite the excellent performances asserted by certain authors, these algorithms remain sensitive to occultation by other components of the road scene. Moreover, color information is highly dependent on the wear and tear of traffic signs, as well as on the illumination conditions. With this in mind, it is difficult to ensure that the results of the segmentation phase are completely reliable [GAV 99]. Characteristics linked to the gradient of the images are also sensitive to image perturbations. Any error at this stage is indeed comparable to a partial occultation of the sought object: the characteristics of the segmented objects are distorted and the classification process becomes much more difficult.
In this chapter, we consider a different application, namely traffic sign inventory from large road scene databases. Such image databases are created by inspection vehicles driving around the French road network. The images are taken by standard numerical cameras under natural illumination conditions. The main area of interest of these inventories ...