研究課題/領域番号 |
20K22383
|
研究機関 | 中央大学 |
研究代表者 |
|
研究期間 (年度) |
2020-09-11 – 2024-03-31
|
キーワード | Localization / Spherical Camera / 360 degree Camera |
研究実績の概要 |
In this year, two different algorithms were implemented and tested in virtual and real environments. One was based on using line information, according to the original proposal. Another one was based on using door information which is usually contained in floor maps of the environment. The method with line information used image gradients. A localization error of around 20cm was obtained in an environment of size 7.6 m × 5.21 m × 2.1 m. This error was also tested under conditions of occlusion and was shown to be robust even if 20% of the image was covered. The method using door information was explored for environments in which line maps were not available, using deep learning based object detection. When tested on real environments, the error was within one meter.
|
現在までの達成度 (区分) |
現在までの達成度 (区分)
1: 当初の計画以上に進展している
理由
In addition to completing the proposed work, another method using object information (doors) contained in floor maps was also explored. This was done by detecting the doors in the image using deep learning-based object detection, and comparing the detections to the floor map. This can allow localization in environments where it is not possible to obtain accurate 3D line information.
|
今後の研究の推進方策 |
In future work, the method using line information will be tested in real-time on various robots. Also, the method using door information will be expanded. Currently, it is in a proof-of-concept stage. A suitable search algorithm will be designed and the accuracy and robustness of the method will be evaluated.
|
次年度使用額が生じた理由 |
The amount for this year will be used to test the original research plan by placing the camera on mobile robots. Performance evaluation will be conducted in real scenarios. Moreover, it will be used to develop the additional algorithm and perform evaluation. Several publications in international conferences and journals are also being planned.
|