Focus First: Coarse-to-Fine Traffic Sign Detection With Stepwise Learning

Liman Liu,Yuntao Wang,Kunqian Li,Jie Li

doi:10.1109/access.2020.3024583

Abstract

Traffic sign detection is a very important part of intelligent assisted driving system. However, with the interference of various target sizes, geometric distortion, occlusion and motion blur, fast and accurate detection on large-size car camera image is extremely hard. To achieve both high efficient and accurate detection, we present a traffic sign detection method within a coarse-to-fine framework, which sequentially detects the targets in grid-level and image-level. We demonstrate that focusing first is a more effective detection strategy for small targets in wide detection space. We propose a target grid prediction network, which is a fully convolutional network for binary classification, to realize rapid coarse localization of the target and effectively guide the clipping and scaling of the target area. With the flexible potential target region extracting strategy, the detecting space can be significantly reduced. At the same time, the correctly extracted local areas for the targets can further facilitate the accurate detection of the subsequent traffic sign detector. In the experiments, our method achieves impressive performance in terms of both efficiency and accuracy. On the challenging Tsinghua-Tencent 100K (TT100K) dataset, our method achieves 20.9 FPS detection speed for 1600 × 1600 images and the F 1 -score of the proposed method for all-scale targets is 91.55.

Highlights

Traffic signs on the roads convey guidance, warnings, restrictions or instructions with words or symbols, which play important roles in regulating the driving behaviour of drivers and ensuring the safety and smoothness of road traffic
Jin et al [20] further improved the accuracy of traffic sign recognition by proposing a hinge loss stochastic gradient descent (HLSGD) method to train convolutional neural networks
EXPERIMENTAL RESULTS OF THE OVERALL ALGORITHM We evaluate the performance of the proposed method on the Tencent 100K (TT100K) dataset and compare it with RefineDet [54], the methods of Zhu et al [59], Li et al [24], Wang et al [49], Liu et al [33] and Noh et al [34], which are the latest advanced approaches for traffic sign detection

Summary

INTRODUCTION

Traffic signs on the roads convey guidance, warnings, restrictions or instructions with words or symbols, which play important roles in regulating the driving behaviour of drivers and ensuring the safety and smoothness of road traffic. Jin et al [20] further improved the accuracy of traffic sign recognition by proposing a hinge loss stochastic gradient descent (HLSGD) method to train convolutional neural networks These algorithms can achieve good detection accuracy and satisfactory speed on common datasets, such as PASCAL VOC [11], MSCOCO [29] and ILSVRC [40] datasets. It can be noticed that, the detection efficiency in traffic sign detection task especially small target detection still needs improving To cover this concern, we propose a coarse-to-fine detection pipeline, whose front module is a grid-level target prediction network. A. TARGET GRID PREDICTION WITH CONVOLUTIONAL NEURAL NETWORK The traffic sign usually only occupies a very small proportion of the image, and it will take a long time to directly detect the high resolution image using the conventional target detection algorithm. 5: Generate a random probability p1 which follows uniform distribution in [0,1]

17: Do random contrast adjustment

Findings

DISCUSSION

CONCLUSION AND FUTURE WORKS