| 講演抄録／キーワード |
| 講演名 |
2025-03-21 11:05
Lightweight Object Detection Model for a CMOS Image Sensor with Binary Feature Extraction
○Keiichiro Kuroda・Yudai Morikaku・Yu Osuka・Ryoya Ieagaki・Kota Yoshida・Shunsuke Okura（Ritumeikan Univ） |
| 抄録 |
（和） |
For the coming of IoT age, we have proposed an object detection system using a CMOS image sensor with binary feature extraction to reduce the power consumption of recognition systems [1]. First, a lightweight deep neural network (DNN) for the feature data is verified based on YOLOv7. The presented model is comparable to the YOLOv7-tiny in the number of parameters and FLOPs, while improving the object recognition accuracy of large objects APL50 by 6.6%. Secondly, we present an on-chip signal processing technique for the CMOS image sensor to extract binary feature data. The simulation results demonstrate that the size of the 1-bit feature data is reduced by 96.0% at the cost of an 8.3% degradation in APL50 compared to the 1-bit RGB color image. |
| （英） |
For the coming of IoT age, we have proposed an object detection system using a CMOS image sensor with binary feature extraction to reduce the power consumption of recognition systems [1]. First, a lightweight deep neural network (DNN) for the feature data is verified based on YOLOv7. The presented model is comparable to the YOLOv7-tiny in the number of parameters and FLOPs, while improving the object recognition accuracy of large objects APL50 by 6.6%. Secondly, we present an on-chip signal processing technique for the CMOS image sensor to extract binary feature data. The simulation results demonstrate that the size of the 1-bit feature data is reduced by 96.0% at the cost of an 8.3% degradation in APL50 compared to the 1-bit RGB color image. |
| キーワード |
（和） |
CMOSイメージセンサー / 特徴量抽出 / 物体検出 / データ量削減 / ランレングス符号化法 / / / |
| （英） |
CMOS image sensor / feature extraction / object detection / data reduction / run-length encoding / / / |
| 文献情報 |
映情学技報, vol. 49, no. 13, IST2025-11, pp. 7-8, 2025年3月. |
| 資料番号 |
IST2025-11 |
| 発行日 |
2025-03-14 (IST) |
| ISSN |
Online edition: ISSN 2424-1970 |
| PDFダウンロード |
|