Post Image Processing for Improving Robustness of CNN-Based Traffic Light Detection on Unmanned Mobile Vehicles

View at publisher Summarise with VIRAT
Authors: Takumi Obayashi, Tatsunosuke Shiota, Yukito Taguchi, Ryuki Tsuruda, Ryosuke Oshima, Shota Koga, Nozomi Tsutsui, Sakuya Yamaguchi, Tan Jin Hung, Hiromu Matsuzaki, Masahiro Kiyota, Ryuto Matsunaga
+5 more Nagisa Yoshikawa, Kento Nakai, Masahiro Nishimura, Taito Manabe, Yuichiro Shibata
Cited by: 0
References: 2

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

This paper discusses post-processing methods to enhance the robustness of traffic signal detection tasks for SoC FPGA-based autonomous mobile vehicles. Although a machine learning object detection approach, such as the use of YOLO, has been proven to be effective, it often faces challenges with false detections in locations without traffic lights and color misclassification, both of which are caused by changes in the operating environment. To solve this problem, we implemented a traffic light and color recognition algorithm in software as a post-processing step, which uses luminance and hue information to complement the YOLO detection. This enables robust recognition that is less affected by lighting conditions or the background, effectively preventing misdetections.

Fields of Study & Subjects

Advanced Neural Network Applications Autonomous Vehicle Technology and Safety Computer Science and Engineering

Publication details

Journal / source
International Conference on Consumer Electronics
Pages
1–5
Publication date
3 February 2026
Language
EN