Kinematic chain motion predictions using results from multiple approaches combined via an artificial neural network

A method, comprising:

• generating, by a plurality of sensor modules attached to a kinematic chain of a user, orientation measurements, each respective sensor module in the plurality of sensor modules having an inertial measurement unit, the respective sensor module being attached to a part of the kinematic chain of the user to track an orientation of the part during a sequence of actions of the user, wherein the kinematic chain includes a plurality of parts connected via joints and at least one part of the kinematic chain has no attached inertial measurement unit;
• receiving, in a computing device coupled to the plurality of sensor modules, the orientation measurements generated by the sensor modules;
• computing, by the computing device using a first technique and based on the orientation measurements generated by the sensor modules, first orientation measurements of the part having no attached inertial measurement unit in the kinematic chain, the first orientation measurements being computed for a plurality of time instances respectively during the sequence of actions of the user;
• computing, by the computing device using a second technique and based on the orientation measurements generated by the sensor modules, second orientation measurements of the part having no attached inertial measurement unit in the kinematic chain, the second orientation measurements being computed for the plurality of time instances respectively during the sequence of actions of the user;
• selecting, by the computing device using an artificial neural network and for each respective time instance in the plurality of time instances, a measurement from:
• a measurement in the first orientation measurements computed using the first technique for the time instance, and
• a corresponding measurement in the second orientation measurements computed using the second technique for the time instance,
• wherein the artificial neural network is trained through:
• generating, using a particular sensor module attached to the particular part of a person, orientation measurements of the particular part at time instances during a sequence of actions of the person;
• and
• comparing the orientation measurements generated by the particular sensor module with first orientation measurements computed using the first technique and second orientation measurements computed using the second technique to identify classifications of whether the first technique or the second technique is more accurate at the time instances during the sequence of actions of the person;
• and
• controlling, by the computing device, a computer model of the kinematic chain of the user using the measurement selected from:
• the measurement in the first orientation measurements computed using the first technique for the time instance, and
• the corresponding measurement in the second orientation measurements computed using the second technique for the time instance.

A method, comprising:

• receiving, in a computing device, orientation measurements generated by a plurality of sensor modules attached to a kinematic chain of a user, each respective sensor module in the plurality of sensor modules having an inertial measurement unit and being attached to a part of the kinematic chain of the user to track an orientation of the part during a sequence of actions of the user, wherein the kinematic chain includes a plurality of parts connected via joints;
• computing, by the computing device using a first technique and based on the orientation measurements generated by the sensor modules, first orientation measurements of a particular part in the plurality of parts connected via the joints, the first orientation measurements being computed for a plurality of time instances respectively during the sequence of actions of the user;
• computing, by the computing device using a second technique and based on the orientation measurements generated by the sensor modules, second orientation measurements of the particular part, the second orientation measurements being computed for the plurality of time instances respectively during the sequence of actions of the user;
• identifying, in the computing device, classifications of whether a measurement in the first orientation measurements computed using the first technique for each respective time instance in the plurality of time instances is more accurate than a corresponding measurement in the second orientation measurements computed using the second technique for the time instance, including:
• generating, using a particular sensor module attached to the particular part, orientation measurements of the particular part at the plurality of time instances during the sequence of actions of the user;
• and
• comparing the orientation measurements generated by the particular sensor module with the first orientation measurements and the second orientation measurements to identify the classifications;
• and
• training, by the computing device using a machine learning technique, an artificial neural network and for each of the plurality of time instances, to classify whether the measurement in the first orientation measurements computed using the first technique for the time instance is more accurate than the corresponding measurement in the second orientation measurements computed using the second technique for the time instance.