Closed ivoflipse closed 11 years ago
This was to be expected given how the parameters are determined. Because the dog is walking along the side of the plate, he's skewing the average size of the paw, which I suspect tricks the euclidean distance check into using smaller distances.
I added these configurable values to merge_contacts in the tracking algorithm.
frame_threshold = np.mean(lengths) * configuration.tracking_temporal
euclidean_distance = np.mean(sides) * configuration.tracking_spatial
average_surface = np.mean(surfaces) * configuration.tracking_surface
This seems to solve the problem we saw in big dogs:
I actually think it won't hurt it small animals either, because the frame_threshold will still filter out any contacts that don't overlap considerably over time.
Here's an example of a rather large dog (Zeff), where one of the toes apparently is quite separated from the rest of this toes and its causing the tracking to make an error.