Video-based hand and ground reaction force determination
A method for determining a hand force and a ground reaction force for a musculoskeletal body of a subject includes obtaining video data for the musculoskeletal body during an action taken by the subject, generating, for each frame of the video data, three-dimensional pose data for the subject based on a three-dimensional skeletal model, and determining the hand force and the ground reaction force based on the three-dimensional pose data. Determining the hand force and the ground reaction force includes implementing a reconstruction of the hand force and the ground reaction force based on the three-dimensional pose data. The method additionally includes applying the three-dimensional pose data, the estimate for the ground reaction force, and the estimate of the hand force, to a neural network or other model to optimize the estimate of the hand force and the estimate of the ground reaction force.
1. A method for determining a hand force and a ground reaction force for a musculoskeletal body of a subject, the method comprising:
obtaining, with a processor, video data for a musculoskeletal body during an action taken by the subject, the video data comprising a plurality of frames;
generating, with the processor, for each frame of the plurality of frames, three-dimensional pose data for the subject based on a three-dimensional skeletal model; and
determining, with the processor, the hand force and the ground reaction force based on the three-dimensional pose data,
wherein determining the hand force and the ground reaction force comprises:
implementing, with the processor, a reconstruction of the hand force and the ground reaction force exerted on the musculoskeletal body during the action based on the three-dimensional pose data, the reconstruction being configured to determine an estimate for the ground reaction force in conjunction with an estimate for the hand force such that the estimate for the hand force is based on the estimate for the ground reaction force; and
applying, with the processor, the three-dimensional pose data, the estimate for the ground reaction force, and the estimate of the hand force, to a recurrent neural network or a computer vision deep neural network to optimize the estimate of the hand force and the estimate of the ground reaction force.
2. The method of claim 1 , wherein the reconstruction is based on a tree model.
3. The method of claim 1 , further comprising providing, with the processor, an output indicative of the hand force to a graphics engine to apply the output to a procedure for rendering an image implemented by the graphics engine.
4. The method of claim 1 , wherein the recurrent neural network comprises a long short-term memory (LSTM) neural network.
5. The method of claim 1 , wherein implementing the reconstruction comprises determining a range for the ground reaction force based on constraints indicative of a set of configurations for the musculoskeletal body.
6. The method of claim 1 , wherein the reconstruction is configured to determine the estimate for the ground reaction force for each frame of the plurality of frames.
7. The method of claim 1 , wherein an optimization routine is configured to determine the ground reaction force for each frame of the plurality of frames.
8. The method of claim 1 , wherein obtaining the video data comprises capturing, with a camera device, the video data.
9. The method of claim 1 , further comprising generating, with the processor, an internal load exerted on the musculoskeletal body based on the ground reaction force and the hand force.
10. The method of claim 1 , further comprising generating, with the processor, an external force exerted by the musculoskeletal body based on the hand force and the ground reaction force.
11. A system for determining a hand force and a ground reaction force for a musculoskeletal body of a subject, the system comprising:
a memory configured to store pose estimation instructions, force reconstruction instructions, and force optimization instructions; and
a processor configured to obtain video data for the musculoskeletal body during an action taken by the subject, the video data comprising a plurality of frames, wherein the processor is further configured to:
upon execution of the pose estimation instructions, generate, for each frame of the plurality of frames, three-dimensional pose data for the subject based on a three-dimensional skeletal model,
upon execution of the force reconstruction instructions, implement a reconstruction of the hand force and the ground reaction force exerted on the musculoskeletal body during the action based on the three-dimensional pose data, the reconstruction being configured to determine an estimate for the ground reaction force in conjunction with an estimate for the hand force such that the estimate for the hand force is based on the estimate for the ground reaction force, and
upon execution of the force optimization instructions, apply the three-dimensional pose data, the estimate for the ground reaction force, and the estimate of the hand force, to a recurrent neural network to optimize the estimate of the hand force and the estimate of the ground reaction force.
12. The system of claim 11 , wherein the reconstruction is based on a tree model.
13. The system of claim 11 , wherein the recurrent neural network comprises a long short-term memory (LSTM) neural network.
14. The system of claim 11 , wherein the memory is configured to additionally store output generation instructions, and
wherein the processor is further configured to, upon execution of the output generation instructions, generate an internal load exerted on the musculoskeletal body based on the estimate of the hand force.
15. The system of claim 11 , wherein the memory is configured to additionally store output generation instructions, and
wherein the processor is further configured to, upon execution of the output generation instructions, generate an external force exerted by the musculoskeletal body based on the estimate of the hand force and the estimate for the ground reaction force.
16. The system of claim 11 , wherein the processor is configured to, upon execution of the force reconstruction instructions, determine the estimate for the ground reaction force for each frame of the plurality of frames.
17. The system of claim 11 , wherein implementing the reconstruction comprises determining a range for the ground reaction force based on constraints indicative of a set of configurations for the musculoskeletal body.
18. The system of claim 11 , wherein the processor is further configured to provide an output indicative of the hand force to a graphics engine to apply the output to a procedure for rendering an image implemented by the graphics engine.
19. The system of claim 11 , wherein the processor obtains the video data from a camera device.
20. The system of claim 11 , wherein the memory is configured to additionally store output generation instructions, and
wherein the processor is further configured to, upon execution of the output generation instructions, generate an ergonomic risk assessment based on the optimized estimate of the hand force and the optimized estimate of the ground reaction force.