|
MIYAMOTO Ryusuke
Department Undergraduate School , School of Science and Technology Position Associate Professor |
|
| Language | Japanese |
| Publication Date | 2019/01 |
| Type | Academic Journal |
| Peer Review | Peer reviewed |
| Title | Comparison of Object Detection Schemes Using Datasets of Sports Scenes |
| Contribution Type | Co-authored (first author) |
| Journal | The Journal of the Institute of Image Electronics Engineers of Japan |
| Journal Type | Japan |
| Publisher | The Institute of Image Electronics Engineers of Japan |
| Volume, Issue, Page | 48(1),pp.144-152 |
| Details | Visual object detection is one of the most difficult tasks in the field of image recognition
but the detection accuracy has been drastically improved by recent machine learning techniques. Two kinds of schemes show good accuracy for object detection: detectors constructed by boosing using decision trees as weak classifiers and detectors based on deep learning. To improve the processing speed of visual object detection based on deep learning without reducing detection accuracy, YOLO adopts grid-based detection instead of sliding windows that requires huge computational costs. In this paper, the detection accuracy of Informed-Filters, Faster R-CNN, and YOLOv2 were evaluated using CG and VS-PETS2003 datasets. Based on the detection results, we discuss about the characteristics of these schemes. |