| Pose estimation versus Kinovea for assessing ankle joint during gait cycle in sagittal plane: Pilot study |
| Paper ID : 1105-ISCSR3 (R1) |
| Authors |
|
Mera Maged Ghatas *1, Mariam Medhat Taki2, Marwa Elkawy Ghait Mohamed3 1Student of Physical Therapy at Sphinx University, Assiut 2Student of Physical Therapy at Sphinx University, Assiut, Egypt 3Assistant lecturer, Department of Physical Therapy for Pediatrics, Faculty of Physical Therapy, Sphinx University, Assiut, Egypt |
| Abstract |
| Background: The motion detection became commonly used in many different fields in life for example: animation and film making, manufacturing and automotive industry, robotics and medical sports rehabilitation. One of the motion detection techniques is pose estimation, the pose estimation is dependent on geometric data which work on the targeted joints in the human surface anatomy then detect them by selecting key points of each joint and connecting them to create a skeleton structure that able to predict the position of each joint in relation to the whole body. It is trained on many data sets that analyze the images and videos for humans with different demographic characteristics in different positions and actions which allow it to able to recognize the action during the motion using neural network mechanism. Purpose: To test the accuracy of automatic detection of pose estimation models compared with Kinovea software. Methods: 10 participants were selected from students of Sphinx University, 8 males and 2 females. Using a stand camera and smartphone, all participants were recorded on video during gait from the sagittal plane and analyzed by Kinovea and a combined model (OpenCV and Mediapipe) to test the accuracy of OpenCV compared with Kinovea. Results: Using a paired t-test in statistics, the finding showed that there is a significant difference between Kinovea and OpenCV in plantar flexion of the left ankle (P value = 0.022699). With the exception of that, all ankle movements and range of motion results are a nonsignificant difference: left ankle (dorsiflexion, P = 0.295666247; ROM, P = 0.15538); right ankle (plantarflexion, P = 0.307683; dorsiflexion, P = 0.173160072; ROM, P = 0.176161). That indicates the inaccuracy of models. Conclusion: The findings referred that OpenCV-Mediapipe can’t be used alternatively due to inaccuracy that needed to more development and modifications for this model. |
| Keywords |
| pose estimation, Kinovea, gait cycle, ankle rangle of motion |
| Status: Abstract Accepted |