Lower Body Detection and Tracking with AlphaPose and Kalman Filters
Fahmy Ferdian Dalimarta, Zainal A. Hasibuan, Pulung Nurtantio Andono, Pujiono Pujiono, Moch Arief Soeleman · 2021
The expansion of computer vision is currently experiencing significant growth, especially in healthcare, security, and other social life aspects. One of the concepts of computer vision advancement is using a regular camera to detect lower body movements, from the waist to toe, to see hyperactivity symptoms in children. This article presents a framework and analysis of lower body movement detection using the estimator alphaPose pre-trained using Halpe 26 keypoints and HOI-Det dataset. The next phase is object tracking using the optical flow of the Pyramidal Lucas-Kanade method. As the final step, we are using the Extended Kalman Filter to determine movement patterns. Finally, the framework will calculate the intensity of the object's movement. Testing this framework was conducted using several videos of human activities. It resulted in detection and movement intensity calculations with an accuracy of 95.17 percent for the lower leg detection, which is better than existing methods.