Objective:
To investigate the feasibility of an automatic extended timed up and go test based on markerless 3D video.
Method:
A 3D camera was used to simultaneously obtain depth and color images of 7 healthy subjects performing 16 timed up and go tests in total. Image processing algorithms were applied on a combination of both image modalities in order to automatically identify several separate components of the timed up and go test: sit, rise, stand, walk, turn, walk back, turn, sit. This detection is based on background segmentation on the combined image modalities and on Hidden Markov Models recognition on the motion vectors of the segmented foreground blobs. For each of these detected components, the duration was automatically calculated and was compared to manually annotated HD video sequences.
Results: An average absolute error of 2.07s (=6.3%) was observed when comparing the total duration of the test as detected automatically compared to manual annotation of the video, resulting in very good correlation (R2=0.98). For individual phases of the test, correlation varied from R2= 0.1 to R2 =0.78.
Conclusion:
These initial experiments show that an automated timed up and go test, with duration information of each of the components of the test can be obtained from markerless motion analysis by means of a 3D camera. This allows for routine assessment and for larger scale studies on the importance of the relative durations of the different components of the test in the assessment of the Parkinson patient.
Original languageEnglish
Title of host publicationXIX World Congress on Parkinson’s Disease and Related Disorders
Publication statusPublished - 13 Dec 2011

    Research areas

  • automated TUG test, 3D camera, mobility, balance, motion analysis

ID: 2197346