We use multiple visual modalities specifically, pressure image and depth image, to predict pressure applied on the human mesh.
Depth image in this context is taken using a depth camera placed above the bed. The pressure image is formed by a pressure mat, which is an array of sensors placed on a mattress. These sensors send an electric signal when force is applied on them, resulting in the output pressure image.
Figure 5: Pressure image forming
Together the depth and pressure image provide complimentary information covering the human resting pose from above and below the person. These representations thus make it possible to predict resting pose and pressure on the human body.
Figure 6: Complimentary Pressure Information in Pressure image and Depth image
Approach 1:
In the first approach, we use the depth image and pressure image to predict the SMPL parameters to form the human mesh. The input pressure image is then mapped to the human body.
Figure 7: Approach 1
Approach 2:
In the above approach, the errors from pose estimation get accumulated in pressure estimation. In this method, we attempt to directly predict pressure.
Figure 8: Approach 2