Moving-object detection method for moving cameras by merging background subtraction and optical flow methods
| Autor: | Takashi Shibata, Takuya Ogawa, Kengo Makino, Shoji Yachida, Katsuhiko Takahashi |
|---|---|
| Rok vydání: | 2017 |
| Předmět: |
Background subtraction
Pixel business.industry Computer science Optical flow High resolution 020207 software engineering 02 engineering and technology Object detection Robustness (computer science) 0202 electrical engineering electronic engineering information engineering 020201 artificial intelligence & image processing Computer vision Artificial intelligence Recall rate business Merge (version control) |
| Zdroj: | GlobalSIP |
| Popis: | This paper presents a novel moving-object detection method for moving cameras. The proposed method merges two scores to detect moving objects more accurately in quasi-real time. We designed two scores, anomaly and motion, for real-time application. The anomaly score is calculated based on a background subtraction and depends on the difference in pixel intensities between the current image and the background model. The anomaly score extract the moving objects pixels with high precision rate, however it tends to result in under-detection (low recall rate) in the object pixels whose intensities are close to the background. The motion score is calculated from a sparse optical-flow, which is based on the short-term tracking results of sparsely sampled points. The motion score can extract the moving object's pixels with high recall rate, even if those pixel intensities are close to the background. However, it tends to result in over-detection (low precision rate) around the object boundaries due to the sparse optical flow. We propose to merge these two complementary scores to compensate for each other, thus it can detect moving objects with robustness and high resolution. Experimental results showed that our proposed method outperforms the conventional methods in terms of F-measure. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|