[home] [Personal Program] [Help]

---

Session: Motor Control II
Session starts: Friday 25 January, 13:00
Presentation starts: 13:30
Room: Lecture room 558

---

Bart Klaassen (University of Twente)
Leendert Schaake (Roessingh Research and Development)
Jaap Buurke (Roessingh Research and Development)
Bart Koopman (University of Twente)
Martijn van Eenennaam (University of Twente)
Hans Rietman (Roessingh Research and Development)

Abstract:
Markerless Motion Tracking (MMT) is a hot topic in surveillance technology, but can also be applied in human kinematics. For example in gait event detection to determine spatio-temporal parameters (e.g. step intervals, heel strike/toe-off postures) and normalization of kinematic or EMG data by detecting individual stride lengths. In daily clinical practice, video images are combined with EMG and/or ground reaction forces for clinical gait assessment of patients. Gait event detection, like identification of Initial Contact Times (ICT), is usually done manually from these video recordings. This can be labor intensive and prone to errors or bias. Therefore the question rises if this detection could be automated for certain events. In this research, main focus is set on ICT extractions from standard video images without any markers on the patient. One of the prerequisites for reproducible gait feature extractions is that the viewing angle from a video camera towards the patient must be perpendicular and approximately constant in order to extract information properly. Therefore, an automatic camera system has been developed, which enables following patients from the side during a walking measurement. OpenTLD [1] has been chosen as the primary platform for the markerless ICT extractions. It can be adjusted and incorporated in additional Matlab scripts that have been developed for extracting ICT. It is based on Tracking-Learning-Detection and outputs x and y coordinates of the tracked item [2]. These scripts can be run in a post-analysis setup. For validating OpenTLD and the combined Matlab scripts, measurements were done with three patients, where the patients had to walk over a 7 meter track followed by the camera system. As a reference, a foot pressure sensor was placed on the calcaneus. Contact times from this sensor were compared with the extracted post ICT results. Differences between the two methods were calculated for slow walking (< 1 m*s^-1), normal walking (about 1.4 m*s^-1) and fast walking (> 2 m*s^-1) speeds. Comparison between post ICT extractions minus foot pressure sensors results in a -0,02 ± 0,10 seconds delay combined over all subjects and all walking speeds. Therefore it shows that as a first attempt, the automatic process for detecting ICT from standard video images using MMT is possible and that the delay is within a certain range which should not affect the outcome for clinical evaluations. Further research is required to identify the cause and significance of this delay, and explore extraction of other gait events.