A Traffic Sign Detection Method based on Saliency Detection
Tianyun Zhao, Qi Liu, Yunpeng Zhang · 2019
As an important research direction of intelligent transportation systems (ITS), road traffic sign recognition research has become increasingly important in autopilot and traffic sign maintenance. The difficulties and challenges faced by research in this area are also increasing, such as feature extraction and regional segmentation. The detection of visual saliency has become a hot topic in the field of computer vision research in recent years. This technology simulates the human visual attention mechanism and can quickly obtain effective information from massive image data. However, in complex situations, due to the effects of changes in the target scale, the complexity of the target contour and background interference, most of the existing algorithms have difficulty achieving a better detection effect, and there is also room for improvement in real-time detection and fusion of saliency maps. Therefore, the detection of visually significant targets is still an important issue that needs to be further studied. Combining the above two points, this paper proposes a traffic sign detection method based on saliency detection. Commonly used saliency detection methods include MBD and MDC. In this paper, the two methods are combined and improved and tested and verified on a data set containing 100 traffic sign pictures. After testing, the average IOU reached 38 percent, the recall rate reached 42 percent and the accuracy rate reached 50 percent. The results show that the proposed method results in significant performance improvement.