Loading...
Please wait, while we are loading the content...
Similar Documents
A study on spatio-temporal CoHOG features for recognition of generic objects in video
| Content Provider | Semantic Scholar |
|---|---|
| Author | Shogo, Nakamura Daisuke, Deguchi Tomokazu, Takahashi Ichiro, Ide Hiroshi, Murase |
| Copyright Year | 2011 |
| Abstract | Recognizing objects in videos is one of the important technologies to search a large amount of videos efficiently on the Web. Recently, generic object recognition has been actively studied for still images, but almost not for videos. As for the generic object recognition in a video, it is important to use both the shape features and the motion features obtained from multiple frames in the video efficiently. In this paper, we propose spatio-temporal CoHOG (Co-occurrence Histograms of Oriented Gradients) features. This is an extension of CoHOG features that provide a high performance for pedestrian detection and others. The spatio-temporal CoHOG features are co-occurrence histograms of oriented spatio-temporal gradients in local regions in a video. In the recognition, a BoF (Bag of Features) representation and a kernel SVM are employed. We conducted an experiment on 1,000 videos including 10 categories collected from the Web. The experimental results showed the effectiveness of the spatio-temporal CoHOG features compared with conventional optical flow features and SIFT features. |
| Starting Page | 1 |
| Ending Page | 6 |
| Page Count | 6 |
| File Format | PDF HTM / HTML |
| Volume Number | 111 |
| Alternate Webpage(s) | http://www.murase.m.is.nagoya-u.ac.jp/~ide/res/paper/J11-kenkyukai-snakamura-1.pdf |
| Language | English |
| Access Restriction | Open |
| Content Type | Text |
| Resource Type | Article |
