Event Detection Based on Noisy Object Information

This paper proposes an event detection method using noisy object information. Some events have a close connection with objects, and the objects related to the event often appear with the event in a video. For example, if an event "Grooming an animal" appears in a video, an animal and people should appear in the video. If we detect the objects that have a close connection with the events, we can detect the events based on the object detection results. However, it is inevitable that the object detector gives false alarms and the object information would be noisy. Thus, we use the information about objects which is robust against false alarms of object detection. In our experiments, we evaluated how the information on objects is effective for event detection of videos. From the results, the proposed method showed better or equivalent detection results than a state-of-theart method in some events, even the object detector gave false alarms.