Moving Image Scene Recognition and Its Application to Highly-Safe Intelligent Systems
Autor: | Kenichi Takada, Michitaka Kameyama |
---|---|
Rok vydání: | 2019 |
Předmět: |
Computer science
business.industry Frame (networking) ComputingMethodologies_IMAGEPROCESSINGANDCOMPUTERVISION Intelligent decision support system Optical flow Convolutional neural network Grayscale Image (mathematics) Support vector machine RGB color model Computer vision Artificial intelligence business |
Zdroj: | 2019 International Conference on Information and Digital Technologies (IDT). |
DOI: | 10.1109/dt.2019.8813701 |
Popis: | High-accuracy recognition of dangerous scenes from the input image information, is essential to realize a highly-safe intelligent system which autonomously makes danger avoidance in the real-world environment where human lives. There are many kinds of dangerous scenes which can be recognized by using moving images such as a violence scene. In this paper, optical flow between adjacent frames for an input moving image is represented by an RGB image. A still image obtained by converting a frame at a certain time into grayscale is superposed with the RGB optical flow image. For the superposed image at each time, a scene recognition result can be obtained by combining a convolutional neural network for extracting features and a support vector machine for classification. By taking majority of these scene recognition results, high-accuracy scene recognition for a moving image can be achieved. |
Databáze: | OpenAIRE |
Externí odkaz: |