Journal: Frontiers in Rehabilitation Sciences

Article Title: Enhancing motion tracking accuracy of a low-cost 3D video sensor using a biomechanical model, sensor fusion, and deep learning

doi: 10.3389/fresc.2022.956381

Figure Lengend Snippet: LSTM RNN with a Kalman filter for the shoulder joint location. The network receives measurements from the sensors and has a recurrent memory of two previous time steps. S: shoulder, E: elbow, W: wrist, α : elbow angle, k: Kinect frame, g: goniometer, i: input, o: output, t: time. One network included the goniometer measurement as input, and the other did not.

Article Snippet: An electronic twin axis goniometer (Biopac Systems, USA) was attached to the participant's arm above the elbow joint using double-sided tape ( ).

Techniques:

Journal: Frontiers in Rehabilitation Sciences

Article Title: Enhancing motion tracking accuracy of a low-cost 3D video sensor using a biomechanical model, sensor fusion, and deep learning

doi: 10.3389/fresc.2022.956381

Figure Lengend Snippet: Experiment environment. The participant is seated on an armless chair with her hand strapped to the manipulator, about 2 m in front of the Kinect sensor. The goniometer and three V120:Trio reflectors are connected to the participant's arm.

Article Snippet: An electronic twin axis goniometer (Biopac Systems, USA) was attached to the participant's arm above the elbow joint using double-sided tape ( ).

Techniques:

Journal: Frontiers in Rehabilitation Sciences

Article Title: Enhancing motion tracking accuracy of a low-cost 3D video sensor using a biomechanical model, sensor fusion, and deep learning

doi: 10.3389/fresc.2022.956381

Figure Lengend Snippet: Top: Boxplot and data points for the absolute residual error with respect to the OptiTrack data of the shoulder, elbow, and wrist joint positions of the Kinect recording and the offline optimization. Bottom: Boxplot and data points of the absolute error of the elbow angle calculated from the Kinect measurements and from the offline optimization with respect to the goniometer measurements. Kinect: Kinect recording error, Opt: optimization error. Red points represent outliers. * p < 0.05, *** p < 0.001.

Article Snippet: An electronic twin axis goniometer (Biopac Systems, USA) was attached to the participant's arm above the elbow joint using double-sided tape ( ).

Techniques:

Journal: Frontiers in Rehabilitation Sciences

Article Title: Enhancing motion tracking accuracy of a low-cost 3D video sensor using a biomechanical model, sensor fusion, and deep learning

doi: 10.3389/fresc.2022.956381

Figure Lengend Snippet: Boxplots and data points of the elbow joint absolute residual error (top), wrist joint absolute residual error (middle) and elbow angle absolute error (for the network trackers), and absolute residual error (for the Kalman filter) (bottom). Kinect: Kinect recording error, Kalman: Kalman filter error, NetG: network with goniometer input error, Net: network without goniometer input error. Black points represent outliers. Blue points represent data from the original position of the Kinect sensor. Red points represent data from the alternative position of the Kinect sensor. * p < 0.05, *** p < 0.001.

Article Snippet: An electronic twin axis goniometer (Biopac Systems, USA) was attached to the participant's arm above the elbow joint using double-sided tape ( ).

Techniques: