Honours Project

Creating animations by using computer vision to extract motion data. Created in Opencv C++ and Unity C#

The project uses OpenCV's computer vision techniques to extract joint angles of animals from videos. The application uses cascading classifiers to detect an animal's legs, in this case, a horse. The video frames that have detected legs are then grey scaled and put through a canny edge detection algorithm. Active contours are then applied to the image that attaches to the edges of the legs. The angles are then calculated using the dot product between the front and end of a contour and taking the start of one contour to the start of the next contour in the series. The angles are then outputted to a file.

A Unity script converts the angles from the output file into quaternions. An animation controller was constructed and attached to the model, this script interpolates from the current angle of the joint to the desired angle, creating animation.

A survey was then carried out that compared the animation and a pose-to-pose created animation to verify if the animation created with the motion data is comparable or even better than the pose-to-pose animation.