Avian Terestrial Locomotion: Evolution, Dynamics and Computer Vision

In the first funding period, with regards to avian hindlimb morphology, we uncovered that the tibio-tarsus and tarsometarsus are morphologically integrated, meaning that these two elements do not vary independently ofone another. Adaptations only occur in the femur and phalanges. Size-related changes in moment of inertia do not reduce the cost of swinging the limb relative to a bird’s size. In the first term of the project, we found that hindlimb kinematics and dynamics do not carry a phylo-genetic signal. Grounded running, a gait prominent in small birds, constitutes a trade-off between stability and maneuverability and that limb geometry at touch down relegates overall limb stiffness. We used high-speed x-ray videos of birds to pioneer new methods for landmark tracking in the first funding cycle, and these methods in turn were used to automate processing of kinematic data. We developed a holistic tracking approach which models all anatomical landmarks in one consistent, probabilistic framework and allows the inclusion of additional knowledge. In experiments based on more than 175,000 manually located ground-truth landmarks, we could show that our automatic tracking approach clearly outperforms standard methods and provides reasonable results for all landmark types. In addition, we developed methods for estimating an object’s center of mass solely using x-ray image intensities and automatically applied this method to data covering more than 7000 individual steps of locomotion. In the projected funding period, we will next test whether proximo-distal gradients in avian long bone density are a means to reduce the cost of oscillating the limbs for locomotion. We will study how hindlimbs with differing specialized functions (e.g., climbing, swimming, etc.) influence muscle activity, kinematics, and kinetics – in particular swing phase dynamics during even and uneven locomotion. For this reason, we will also analyze non-cyclic motion of our previously sampled avian species. With regard to computer vision, we will develop a method for landmark tracking for these non-cyclic movements and for objects moving in and out of the camera’s view, as well as methods for landmark tracking that incorporates anatomical and biomechanical knowledge. The new methods for motion analysis will have applicability to biomechanics and future questions beyond the scope of this project.

Antragsteller Prof. Dr.-Ing. Joachim Denzler
Lehrstuhl Lehrstuhl Angewandte Informatik
Drittmittelgeber DFG - Deutsche Forschungsgemeinschaft
Laufzeit Mai 2014 - April 2016
Stellen 1
Besetzung N.N.
Projektart Einzelprojekt
Link zur Webseite