Traffic Sign Detection is one of the most important techniques for intelligent vehicles, and it’s an essential part for autonomous vehicles and driver assistance. Traffic Sign Detection is still an unsolved problem since there are too many challenges, such as weather conditions and lighting conditions. Robustness is always an issue.
There are three main parts in Traffic Sign Detection: Detection, Classification, and Tracking. In our work, we focus on the detection part and incorporate with human perception. The literatures show that color and shape are two essential factors for human to detect traffic signs. In our work, we extracted the color information of traffic signs, and designed a perception-based fusion module which takes spatial and temporal connection into consideration. This module will remove most of the false alarms and keep the objects that have all the color bases in the sign we want to detect. Our next step is to incorporate the shape component to make the whole algorithm more robust.
Traffic sign detection by color-component extraction and perception-based fusion. The first three quarters show the most important three color-components (e.g. red, black, white for no-parking signs), and the last quarter shows the pixel-based detection results.