Racing Bib Number Localization Based on Region Convolutional Neural Networks
Abstract— This work aims to apply region convolutional neural network technique for detecting racing bib numbers on marathon images. Basically, marathon images are captured in the outdoor environment. The images are often ruined because of environment variations such as illumination, occlusion, and complex background. Transfer learning technique is applied for detecting a racing bib number on marathon images. Transfer learning algorithm firstly loads pre-trained model that was already trained on Cifar-10 image dataset. After that, racing bib number images (training images) are trained together with the pre-trained model into two classes: racing bib number class and non-racing bib number class. Finally, the trained model from the previous step is used to detect a racing bib number in marathon images. The results were reported to overtake previous research and F-measure was at 90 % on 400 marathon images.
Index Terms— racing bib number, localization, deep learning, region convolutional neural network.
Noppakun Boonsim, Saranya Kanjaruek
Faculty of Applied science and Engineering, Khon Kaen university, THAILAND
Cite: Noppakun Boonsim, Saranya Kanjaruek, "Racing Bib Number Localization Based on Region Convolutional Neural Networks," Proceedings of 2018 the 8th International Workshop on Computer Science and Engineering, pp. 293-297, Bangkok, 28-30 June, 2018.