The Robust Recognition of Traffic Signs from a Moving Car

The robust and reliable recognition of traffic signs from a moving car is investigated as a specific example of the general ambitious goal of object recognition in natural surroundings. The newly proposed method of hierarchical spatial feature matching is employed, based on a pyramid representation of the scene and its local orientations. The worked example of designing a suitable template for “right-of-way” -signs (diamonds, rotated squares) illustrates some general principles of hierarchical feature matching. Hardware considerations indicate that the problem can be solved in real-time with a data-flow architecture using commercially available matching ICs. The performance with video imagery taken from a moving car (day and night, city and country scenes) is reported and some practical problems encountered are discussed. It is concluded that this non-AI based approach to traffic sign recognition is reliable, simple, fast and performs very well in real-life situations.