Markerless motion capture of animate subject with prediction of future motion
A motion prediction system for predicting the motion of a random animate subject. A first neural network is a markerless motion capture network, trained to receive video data of the subject and to process the video data to generate a time sequence of musculoskeletal motion capture data. A second neural network is a motion prediction network, trained to receive the musculoskeletal motion capture data and to process the data to generate a prediction of the subject's location based on position change in position of joints and/or muscles.
1. A method of predicting a motion of a body of a random animate subject, comprising:
selecting a set of musculoskeletal segments internal to the body;
wherein the set of musculoskeletal segments represent include at least one or more joints within the body that are not externally visible;
wherein the musculoskeletal segments represent portions of the body whose past motion is predictive of the motion of the entire body;
training a first neural network to receive a time sequence of markerless video data of the subject and to process the video data to generate a time sequence of musculoskeletal motion capture data of the subject;
the musculoskeletal motion capture data represents motion of the musculoskeletal segments;
training a second neural network to receive the musculoskeletal motion capture data and to process the data to generate a prediction of the subject's location based on position change of the set of musculoskeletal segments;
capturing video data of the subject;
delivering the video data to the first neural network;
delivering the musculoskeletal motion capture data to the second neural network; and
using the output of the second neural network to predict a future path of the subject.
2. The method of claim 1 , wherein the musculoskeletal motion capture data is at least a portion of the subject's joints and wherein the second neural network is trained to track the movement of the subject's pelvis.
3. The method of claim 1 , wherein the musculoskeletal motion capture data is at least the subject's skull and wherein the second neural network is trained to track the movement of the subject's head pose.
4. The method of claim 1 , wherein the position change is based on x-y position data.
5. The method of claim 1 , wherein the second neural network is trained to predict the subject's position based on less than three seconds of video data.
6. The method of claim 1 , wherein the second neural network is trained by weighting later portions of the motion capture data more than earlier portions of the motion capture data.
7. The method of claim 1 , wherein the second neural network is trained by receiving training data representing changes in balance, changes in center of gravity and/or muscle flexion of subjects depicted in the training data.
8. The method of claim 1 , wherein the second neural network is a stacked long-short term memory network.
9. A motion prediction system for predicting a motion of a body of a random animate subject, comprising:
a first neural network trained to receive a time sequence of markerless video data of the subject and to process the video data to generate a time sequence of musculoskeletal motion capture data of the subject;
wherein the musculoskeletal motion capture data sequence represents a set of musculoskeletal segments internal to the body;
wherein the set of musculoskeletal segments represent include at least one or more joints within the body that are not externally visible; and
wherein the musculoskeletal segments represent portions of the body whose past motion is predictive of the motion of the entire body;
a second neural network trained to receive the musculoskeletal motion capture data and to process the data to generate a prediction of the subject's location based on position change of the set of musculoskeletal segments.
10. The motion prediction system of claim 9 , wherein the musculoskeletal motion capture data is at least a portion of the subject's joints and wherein the second neural network is trained to track the movement of the subject's pelvis.
11. The motion prediction system of claim 9 , wherein the musculoskeletal motion capture data is at least the subject's skull and wherein the second neural network is trained to track the movement of the subject's head pose.
12. The motion prediction system of claim 9 , wherein the position change is based on x-y position data.
13. The motion prediction system of claim 9 , wherein the second neural network is trained to predict the subject's position based on less than three seconds of video data.
14. The motion prediction system of claim 9 , wherein the second neural network is trained by weighting later portions of the motion capture data more than earlier portions of the motion capture data.
15. The motion prediction system of claim 9 , wherein the second neural network is trained by receiving training data representing changes in balance, changes in center of gravity and/or muscle flexion of subjects depicted in the training data.
16. The motion prediction system of claim 9 , wherein the second neural network is a stacked long-short term memory network.
17. An on-board motion prediction system for predicting a motion of a body of a pedestrian in or along a roadway on which a vehicle is traveling, comprising:
at least one video camera for capturing video data of the pedestrian;
a first neural network trained to receive a time sequence of the video data and to process the video data to generate a time sequence of musculoskeletal motion capture data of the subject;
wherein the musculoskeletal motion capture data is a time sequence representing a set of musculoskeletal segments internal to the body;
wherein the set of musculoskeletal segments represent include at least one or more joints within the body that are not externally visible; and
wherein the musculoskeletal segments represent portions of the body whose past motion is predictive of the motion of the entire body;
a second neural network trained to receive the musculoskeletal motion capture data and to process the data to generate a prediction of the subject's location based on position change of the set of musculoskeletal segments; and
a collision avoidance process operable to determine if there is a likely collision with the pedestrian.
18. The motion prediction system of claim 17 , wherein the musculoskeletal motion capture data is at least a portion of the subject's joints and wherein the second neural network is trained to track the movement of the subject's pelvis.
19. The motion prediction system of claim 17 , wherein the musculoskeletal motion capture data is at least the subject's skull and wherein the second neural network is trained to track the movement of the subject's head pose.
20. The motion prediction system of claim 17 , wherein the position change is based on x-y position data.