2017 Fiscal Year Research-status Report
Real World Heart Rate Estimation from Video
| Project/Area Number |
17K12709
|
|---|
| Research Institution | Saitama University |
|---|
Principal Investigator |
Lam Antony 埼玉大学, 理工学研究科, 助教 (50744124)
|
|---|
| Project Period (FY) |
2017-04-01 – 2019-03-31
|
| Keywords | video ppg / computer vision / heart rate / cardiac estimation / remote measurement |
|---|
| Outline of Annual Research Achievements |
We aim to take videos of faces and effectively estimate heart rates (HR) for real-world settings. We first built our own video dataset in a less controlled setup than the Mahnob-HCI dataset. Then we improved over our previous algorithm (ICCV2015). Our past algorithm was effective against illumination changes but less effective with complex motions of the face (e.g. pose changes, large translations). We developed an improved algorithm that uses triangulation to track face regions over pose changes more effectively. We have even tested on more challenging cases such as YouTube videos. We first uploaded self-made videos with complex motions to YouTube with known HRs and verified accurate HR estimation. We have also conducted tests on estimating emotions from our video-based cardiac algorithm.
|
| Current Status of Research Progress |
Current Status of Research Progress
2: Research has progressed on the whole more than it was originally planned.
Reason
We have built our own dataset as indicated and we are improving our algorithm to robustly estimate heart rate (as we proposed). We have even found the new algorithm is effective on YouTube videos (with ground truth HR for validation). We are also doing a lot of investigations on its effectiveness in affective computing applications. These tests are being conducted on the dataset we built (where people watched different movie clips). We are also doing similar tests on real vlogger movie trailer reaction videos with promising results. The main drawback of our algorithm is that it is computationally expensive and therefore slow.
|
| Strategy for Future Research Activity |
We will work on improving the robustness of the algorithm and also its speed. We will also continue investigating emotion recognition using our algorithm, which is able to give the full cardiac pulse signal and not just the HR. By reading emotion from a physiological signal, we may be able to read into human emotional reactions at a deeper level than by just facial expressions alone.
|
| Causes of Carryover |
We plan to use the funds for travel, equipment, and collection of more data for next fiscal year. We expect that all funds will be used next fiscal year.
|
