Morphological Validation of Animal-Skeleton Estimation Result

In this study, we use a morphological analysis-based method for the time sequence of animal-skeleton estimation, which helps recognize bad estimation and split the video accordingly. Signed areas of four polygon regions indicated by the animal’s four legs are used to describe the motion pattern of an animal’s legs. The nine signed areas have a significant change pattern under the given action, while unconventional abrupt changes in the time series indicate serious errors in skeleton identification, such as identifying the front left leg of the animal as the front right leg. Therefore, we can analyze these area changes to identify the wrong frames and use them as a basis for segmenting long segments of video. The experimental result verifies the validation of the analysis method.