2006 Fiscal Year Final Research Report Summary
Image Processing Method for Realizing Privacy Protection and Object Recognition for Fixed Monitoring Camera Systems
| Project/Area Number |
17560329
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
Communication/Network engineering
|
|---|
| Research Institution | Tokyo University of Agriculture and Technology |
|---|
Principal Investigator |
KITAZAWA Hitoshi Tokyo University of Agriculture and Technology, Institute of Symbiotic Science and Technology, Professor, 大学院共生科学技術研究院, 教授 (60345329)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
TANAKA Toshihisa Tokyo University of Agriculture and Technology, Institute of Symbiotic Science and Technology, Associate professor, 大学院共生科学技術研究院, 助教授 (70360584)
|
|---|
| Project Period (FY) |
2005 – 2006
|
| Keywords | Image processing / Privacy protection / Fixed monitoring camera / Watermarking / JPEG compression |
|---|
| Research Abstract |
Public video surveillance by security camera is becoming more and more important for investigation and deterrent of crimes. Therefore we propose a new framework for security camera system which solves the privacy protection problem. The privacy protection is realized by masking or erasing moving objects in the monitored images. The original moving object images are encrypted and watermarked into the masked image. When it is necessary, the original image can be reconstructed using a special viewer with a password.
1. Scrambling and erasing:
First, generates a masked image in which moving objects are scrambled or erased. By the scrambling, all pixels in a moving object are permuted randomly, and by the erasing, all pixels in a moving object are replaced by the corresponding pixels in the background image. The masked images show "what is happening" but do not show "who is there."
2. Generating the masked images:
The extracted moving object is compressed by JPEG, and embedded into masked image. The masked image and the moving object are compressed by high compression level and low compression level, respectively. As a result, we decrease data size by approximately 25% compared to when low compression level is used for both images. Moreover, by utilizing common background image during N frames, data size was further reduced by approximately 50%.
3. Data encryption and reconstruction:
The extracted moving object is encrypted by the advanced encryption standard (AES). Moreover, we can selectively reconstruct one of the masked objects while keep the other objects masked.
4. Improvement of moving object extraction:
The proposed method is strongly influenced by the moving object extraction. We proposed a moving object extraction method which uses normalized vector distance (NVD) of RGB values. We state this method can realize shadow detection under direct sunlight, and decrease the extraction error of moving object.
|
