Traffic Sign Recognition Based on Up-sampling Convolution
Abstract— This paper presented a method makes traffic sign recognition faster and more accurate. Traditional faster detectors are limited by their accuracy and are not sensitive to small objects, in the area of self-driving, it has some inconspicuous but important object of concern, such as traffic sign. We noticed that most traffic signs in dataset is small and easily to confuse with complex backgrounds. In this situation, after a series of convolutional layers, some of these traffic signs can’t be detected or classified correctly, and the problem of neglect happens a lot. In order to settle this problem and optimize the result, we simplified the SSD structure and introduced an up-sampling structure to make the geometric details of small objects distinctly. This method significantly improved the result of recognition, we got 97.6% mAP on The German Traffic Sign Benchmark with 96 × 96 input and SSD300 has 79.7% mAP on same dataset.
Index Terms— traffic sign recognition, up-sampling, small objects
Yitian Lu, Ping Jiang
School of Electrical Engineering, Nantong University, CHINA
Yitian Lu, Shun Nishide, Xin Kang and Fuji Ren
Faculty of Engineering, Tokushima University, JAPAN
Cite: Yitian Lu, Ping Jiang, Shun Nishide, Xin Kang and Fuji Ren, "Traffic Sign Recognition Based on Up-sampling Convolution," Proceedings of 2019 the 9th International Workshop on Computer Science and Engineering, pp. 136-142, Hong Kong, 15-17 June, 2019.