Taiwan Stop Sign Recognition with Customize Anchor
Christine Dewi, Rung-Ching Chen, Yanting Liu, Ye-Shan Liu, Ling-Qi Jiang
Abstract
Traffic signs recognition (TSR) is the main issue for a driver assistance system as it has a dual role to control the road traffic as well as warning and guiding the driver. This paper focus on the Taiwan stop sign, we collect the image and build the dataset by our self. Our aim is real-time detection at a much less processing time. In this work, we will analyze the importance of anchor calculation using k-means and original YOLO V3 for Taiwan stop sign detection and recognition. We conduct some experiments with a different setting that could be seen in Table 1. The experimental results show that our proposed model experiment 3 has a better result compared to other experiment settings. The trend of accuracy increases and the time is decreases from experiment 1, experiment 2 and experiment 3. Anchor recalculation based on our dataset is very important and it is proved in our experiment.