Fig. 1. Illustration of multiple people recognition and positioning with multiple cameras based on “vibration signals” from skin surfaces
Fig. 2. Principle derivation of difference and synchronization of “vibration signals” from skin surfaces
Fig. 3. Validation of synchronization and difference. (a) Experimental scene 1; (b) experimental scene 2; (c) example of IPPG signal acquisition for different skin areas of the same object under the same camera; (d) IPPG signal acquisition for different skin areas of different objects under the same camera
Fig. 4. IPPG signal acquisition for the same skin area of different objects under multiple cameras
Fig. 5. Flow chart of human body recognition and positioning system based on multiple cameras
Fig. 6. Acquisition of “vibration signals” from human skin surfaces
Fig. 7. Positioning with three cameras. (a) Top view of experimental scene; (b) principle diagram of positioning
Fig. 8. One frame of acquired video with three cameras and tracked skin areas
Fig. 9. IPPG signals acquired by camera 3
Fig. 10. IPPG signals acquired by cameras 2 and 3
Fig. 11. Multiple coordinate calculation results of objects A and B
Fig. 12. Experimental scene of 5 objects for recognition and positioning with 3 cameras. (a) Experimental scene for multi-object recognition and positioning; (b) top view of scene; (c) one frame of acquired video by three cameras and skin areas
Fig. 13. Schematic of accuracy area distribution of camera
Fig. 14. Coordinate calculation results of 5 objects
Method for people recognition and positioning | Based on face recognition | Based on “vibration signals” from skin surfaces |
---|
Matching feature | Image feature (geometric feature) | “Vibration signal” from skin surfaces (light intensity change of skin surfaces caused by blood volume change over time) | Area of interest | Face (local area) | Bare skin (whole body area) | Requirement for areas of interest | Light, facial expression, posture, angle, occlusion, image resolution | Light | Safety | Facial features easy to copy or change | “Vibration signal” from skin surfaces related to heartbeat and not easily changed | Database | Need for database | No need for database, instant matching | Data to transfer | Image | One-dimensional time series |
|
Table 1. Method comparison for multiple people recognition and positioning with multiple cameras
Cosine after detrend | A-right arm | A-left arm | B-right arm | B-left arm | A-head | B-head |
---|
A-right arm | 1.0000 | 0.6797 | -0.1338 | 0.1421 | 0.6215 | 0.0128 | A-left arm | | 1.0000 | 0.0467 | 0.1422 | 0.6642 | 0.0084 | B-right arm | | | 1.0000 | -0.7226 | -0.1806 | -0.4908 | B-left arm | | | | 1.0000 | 0.2061 | 0.6557 | A-head | | | | | 1.0000 | 0.0102 | B-head | | | | | | 1.0000 |
|
Table 2. Similarity of IPPG signals acquired by camera 3
Cosine after detrend | Cam2-A-left arm | Cam2-B-left arm | Cam2-A-right arm | Cam2-B-head | Cam2-A-head |
---|
Cam3-A-right arm | 0.493862 | -0.331650 | -0.424230 | 0.231214 | -0.40585 | Cam3-A-left arm | 0.581386 | -0.227610 | -0.754750 | 0.242505 | -0.67486 | Cam3-B-right arm | -0.139560 | 0.337164 | 0.180864 | -0.394820 | -0.02642 | Cam3-B-left arm | 0.349941 | -0.403400 | -0.449830 | 0.406926 | -0.26560 | Cam3-A-head | 0.537608 | -0.180590 | -0.703850 | 0.242280 | -0.43896 | Cam3-B-head | 0.085119 | -0.468560 | -0.232320 | 0.436905 | -0.17860 |
|
Table 3. Similarity of IPPG signals acquired by cameras 2 and 3
Sample | Skin area [format: object number-skin number (camera number)] |
---|
Cluster 1 | 2-1(Cam1),2-2(Cam1), 4-1(Cam1), 2-2(Cam2), 2-2(Cam3), 2-1(Cam3), 1-2(Cam2), 4-2(Cam2), 1-1(Cam1), 4-1(Cam2) | Cluster 2 | 3-1(Cam2), 1-1(Cam3), 1-2(Cam1) | Cluster 3 | 3-2(Cam1), 1-1(Cam2), 3-2(Cam2), 3-2(Cam3), 3-1(Cam1), 3-1(Cam3) | Cluster 4 | 4-1(Cam3), 4-2(Cam3) | Cluster 5 | 5-1(Cam2), 5-2(Cam2), 5-1(Cam3), 5-1(Cam1) |
|
Table 4. Classification results of similarity comparison and normalization
Class | Precision | Recall rate |
---|
Class 1 | 0.67 | 0.40 | Class 2 | 0.50 | 1.00 | Class 3 | 0.83 | 0.83 | Class 4 | 1.00 | 0.40 | Class 5 | 1.00 | 1.00 | Average | 0.80 | 0.73 |
|
Table 5. Precision and recall rates of classification results of similarity comparison and normalization