OMNIDIRECTIONAL CAMERA-BASED TRAFFIC LIGHT DETECTION USING DEEP LEARNING APPROACH
- Delivery
- Available on this site
- Format
- Price
- Non-members (tax incl.):¥1,100 Members (tax incl.):¥880
- Publication code
- 20219003
- Paper/Info type
- Other International Conferences
- Pages
- 1-3(Total 3 p)
- Date of publication
- Sep 2021
- Publisher
- JSAE
- Language
- English
Detailed Information
Author(E) | 1) Yasuyuki Ishida, 2) Toshio Ito |
---|---|
Affiliation(E) | 1) Shibaura Institute of Technology, 2) Shibaura Institute of Technology |
Abstract(E) | In recent years, there has been a growing need for automated driving due to issues such as urbanization and aging populations. Autonomous mobility needs to cross the road safely, and one of the most important objects is the traffic light. Therefore, we detect traffic lights, including their color information. It is suitable for automated vehicles since the omnidirectional camera has a wide field of view. However, the omnidirectional image is different from an ordinal rectangle image, that the objects are arranged in a circle and the spatial resolution in the radial direction is non-constant. Thus, object detection requires different feature information rather than ordinally rectangular images. We divide the omnidirectional image into four parts so that the rotations are in the same direction. Then, we encoded the divided images and input them into the neural network to improve the accuracy. |