They also investigate the sharing of parts [101], and not surprisingly, non-shared model is better, albeit slightly, in both pose and landmark accuracies.
For task-oriented performance, one can measure the impact of the landmarking accuracy on the performance scores of a task.
The landmarking accuracy on the performance of the registration, expression classification, gesture recognition and AAM fitting algorithms, respectively, would be goal-oriented measures of landmarking.
However, even excluding eyebrow outer corners and evaluating the performance over the remaining 15 landmark points did not change the performance: STASM, AAM, FFPD, BORMAN, and ZhuRamanan. Figure 7 Comparison of landmarking accuracy due of different landmark of types (BioID database).
(a) Detection rates of the seven different methods on the m 7 landmark set; (b) Detection rates of the five different methods (FFPD, AAM, STASM, BORMAN, ZhuRamanan), on the m 17 landmark set. Figure 9 Comparison of landmarking accuracy for different facial expressions (CK+ database).
In the future, with the selection of smart sensors for obtaining the position of the robot, such as a higher resolution encoder or multiple sensor fusion approaches in the prototype, we can improve the landmark detection accuracy and explore outdoor environments.
Future studies involving the technique could focus on finding how much additional landmarks increase accuracy.
These were used as the ground truth reference point for all subsequent training and test sets for evaluations of automated facial landmark annotation accuracies.
For the landmark task collected accuracy and RTs (for correct trials only) were recorded during the scanning session.
