RECOGNITION OF BLUE TRAFFIC SIGNS ENHANCED BY AGGREGATING FRAGMENTARILY SEGMENTED SIGN IMAGES

摘要

This paper proposes a new algorithm for aggregatingrnfragmented images to enhance the recognition of bluerntraffic signs designating directions. The system isrndesigned for the assistance of safe driving. Subspacernmethod and Histogram of Oriented Gradients (HOG)rnfeatures are used for classification. More than onernsubspace for each arrow class is used to cover complexrndistributions of arrow patterns.rnThe system consists of several processing stepsrnincluding labeling the blue objects in each frame,rnsegmenting them from the background and classifyingrnthem by subspace method. One of the difficulties in thisrnprocess is the fragmented detection of single signs.rnDirection signs consist of a white arrow surrounded by arnblue background. Due to motion blur or some lightingrnconditions, the white area can spread into the blue arearnand blue background of the sign can be disconnected afterrnthe labeling step. To aggregate the broken backgrounds ofrnthe signs a new algorithm is designed by taking intornaccount the shape information of the signs and arrows.