[20a-P01-1~21] 3.7 Optical measurement, instrumentation, and sensor

[20a-P01-19] Effect of Image Distortion on Convolutional Neural Network Recognition

〇Daichi Sone¹, Kento Kumagai¹, Wei Dong¹ (1.NUT)

Keywords:

360-degree cameras,deep learning,3D measurement

This study aims to develop an omnidirectional object detection and positioning system by combining a 3D measurement system using an omnidirectional camera with YOLO v5. We attempted outdoor object detection and created a custom dataset for cigarette butt detection to verify detection accuracy. Specifically, distorted fisheye camera images were used for real-time object detection through deep learning. This demonstrates the potential for expanding the application range of high-performance omnidirectional cameras.