A Traffic Sign Text Detection System for Pratical Natural Scenes
Zhongrong Zuo, Pengtao Yang · 2018
Up to now, object detection capabilities have recently made extraordinary leaps, thanks to the application of deep learning in the field of computer vision. Currently, most of the object detection algorithms are variations of RPN (Region Proposal Network), and have basically met the requirements of industrial applications. However, in the specific business deployment, there will be many problems, such as more complex actual images, large number of small objects, poor-quality images and so on. To solve these problems, some well-designed operations need to be implemented before the algorithm can be deployed successfully. Aiming at many problems in the actual deployment of traffic sign text detection, this paper proposed a multi-stage detection pipeline, which solves many problems of traffic sign text detection in the actual road conditions. Finally, the AP (Average Precision) in our own test set is 0.9918 (IOU=0.5), which meets the practical requirements.