Scene Recognition Method by Bag of Objects Based on Object Detector
Shuichi Masuda, Yuki Kaeri, Yusuke Manabe, Kenji Sugawara · 2018
The scene recognition is one of the most important tasks for estimating ambient attributes from the image data. The ambient attributes are scene traits for specifying the meanings of human activities. We consider that the estimation of ambient attributes is required for understanding human activities in various scenes. Thus this paper proposes a novel method of the scene recognition by the object detector. The proposed method estimates a scene using a histogram of objects, which is called Bag of Objects, based on the result of object detection. To evaluate the proposed method, a simulation experiment has been done with a lot of image data we collected from the web. As the result, we show that the average of scene recognition accuracy is 0.58 for 26 scene categories.