Subsign detection with region-growing from contrasted seeds

Speed limit determination systems for cars based on vision are more and more developed. Roadsign detection is nowadays a well managed problem. However, in some situations this information is not sufficient to know the speed limitation. Restrictions are sometimes applicable and specified by subsigns. These small rectangles often provide essential information about the applicability scope (vehicle type, condition, lane, etc.) of speed limits. We present an approach of subsign localization based on region growing with an initial step of seed selection using morphological reconstruction. A comparison is also performed with three other techniques based on edge, color and graph on two databases gathering French and German subsigns. The obtained subsign correct detection is above 65%.