A road sign detection method, comprising: a scene modeling step, respectively determining the allowable range of each color according to sample images of a plurality of scene categories; a color filtering step, determining the scene category of the image to be processed, and using the allowable range of each color , to filter the image to be processed, to obtain the respective binarized images of the image to be processed about each color; the geometric feature filtering step is to perform connected domain analysis on each binarized image, and obtain the connected domains in the binarized image Geometric features, exclude the connected domains whose geometric features do not meet the specifications of road signs, and the remaining connected domains are candidate road signs; the shape detection step, according to a limited number of predetermined shapes, detect the shape of the candidate road signs; the recognition step, for the shape-determined candidate The road sign identifies the attribute of the candidate road sign according to the degree of similarity between the internal features and the standard internal features. A road sign detection device is also correspondingly provided.