Video description method for action detection based on part-of-speech information
The invention belongs to the field of computer vision, and provides a video description method for action detection based on part-of-speech information, which comprises the following steps of: (1) extracting video features, (2) extracting scene features, (3) constructing a progressive part-of-speech...
Saved in:
| Main Authors | , |
|---|---|
| Format | Patent |
| Language | Chinese English |
| Published |
09.08.2022
|
| Subjects | |
| Online Access | Get full text |
Cover
| Summary: | The invention belongs to the field of computer vision, and provides a video description method for action detection based on part-of-speech information, which comprises the following steps of: (1) extracting video features, (2) extracting scene features, (3) constructing a progressive part-of-speech decoder, (4) constructing a description generation decoder, and (5) generating a video description statement. According to the method, a part-of-speech information detection module composed of a graph convolutional neural network and a progressive part-of-speech decoder is utilized to judge the global dependency relationship among a plurality of objects in a video so as to determine subject objects, and more accurate part-of-speech information is obtained by combining subject categories and video dynamic feature prediction actions. And finally, a decoder is generated through grammar-guided description, and a more accurate scene description statement is generated.
本发明属于计算机视觉领域，提供了一种基于词性信息的动作检测的视频描述方法，包括：步骤(1)提取视频特征 |
|---|---|
| Bibliography: | Application Number: CN202210478744 |