The experience with human animation (ActIPret project) shows us that the realistic effect can often be achieved by tuning the limiting parameters of the joints. Therefore if we implement the parameters that limit the motion range and mobility of the joint to imitate the reality, we might achieve the realistic effect of the animation without the loss of universality.
There are two types of joint motion parameters. One that sets boundaries to the space of possible states, in case of rotational joint this means upper and lower rotation limit. The other type basically defines priority of certain states of the joint. The problem is how to implement these parameters and how to fill them with realistic values.
The area of inverse kinematics based human animation does not promise any scientific contribution with significant novelty. We list it as a research area because it is necessary to master it for further research on joint parameters.
Methods to generate new animations from examples have been subject of research. They produce new motions and it is necessary to evaluate if these motions are physically possible. Our model of the joint could be used for this purpose.
Our interest in generating new motions is motivated by the old master thesis project (Teaching Tennis in Virtual Environment). Interpolation from examples seems to be a reasonable way of motion data aquisition for this project.
My thesis on shoulder complex animation was succesfully defended in June 2009. The whole title was Computer animation and shoulder complex modelled with respect to real anatomy. Shoulder complex served as a testbed for development of general joint model for use in any kinematic chain of human body.
The shoulder was modeled as an open kinematic chain of 3 joints. Data to animate this chain were obtained by optical motion capture. The chosen animation technique was CCD iterative method of inverse kinematics. This method does not work with orientation data, only position, which suits the optical MoCap input.
The main contribution of the thesis was the secondary objective, the adaptation of CCD animation method for work with more end-effectors. Problems of shoulder complex motion capture were identified and approaches to IK animation were suggested.
The model of human joint was based on motion constrains expressed by animation parameter usually called "stiffness". We turned it into a function rather traditional static parameter. We identified several factors influencing it. Each of them was expressed as a simple function ranging from 0 to 1. The results were put together by fuzzy logic operators.
The abstract of thesis is here to download. The full text of thesis is available, I can send it in response to your e-mail. The language of both documents is czech.
This was commercial project that was running at CMP. I participated as a programmer with a task to develop and implement the nearest neighbor search algorithm. This algorithm was based on the k-D tree data structure built over the data and traversed using the best-bin-first strategy. Although we were planning to implement published approximative algorithm, we got 100% match between our results and brute-force algorithm on all our testing data.
The purpose of nearest neighbor search was in entropy computation which was used to get mutual information between two images. The project is finnished.
EU IST-2001-32184. The full name of the project was Interpreting activities of expert operators for teaching and education. Its goal was to develop methodology to observe, learn and interpret human activity, typically a manipulation task. The learned activity was stored in a form of generalized activity plan for later use, which could be a replay or more importantly a recognition.
There were five partners in four countries participating on the project. The project was running within the timespan of November 2001 and October 2004 and was finished with very positive reviews.
Part of the project was the module for presentation of results in virtual environment which was one of the responsibilities of CMP, Prague. The very nature of the task was a conflict between the tendency to generalize inherent to the artificial inteligence methods used by the ActIpret system (increase of level of abstraction) and and the principle of explanatory visualization which basically create the instance of the visualized phenomenon (decrease of level of abstraction). The work on the VR presentation module helped to identify the problems in this area.
The design of VR presentation module was published in the SCCG'05 paper and the project's final deliverable. In the publication list on this page there are total of three ActIPret deliverables and two papers (SCCG'05 and VIIP'03) that are concerned with this project.
This was an internal CMP project in cooperation with Dr.Wittner from Laboratoty of Cognitive Physiology at the Institute of Physiology (Charles University). We were asked to develop a method to track a laboratory rat in an experimental setup. I implemented a tracker in Matlab. The results were quite satisfactory. The implementation of the full version was decided to be assigned as a master thesis. To my knowledge, no student has been working on it.
At CMP, there is an official web of this project including some results.
My master thesis. Moreover it was published at SCCG'02 conference. The aim of the project was to inquire the possibilities of using virtual reality tools (such as human animation) in sports training, chiefly in teaching precise techniques.
The most important impact of this project was the initial gathering of experience with human animation. We have identified some interesting research project related to the topic. The main problem concernes the aquisition of the data that would be visualized and analysed.
The project was finished, but I would rather consider it "sleeping", as it still can provide framework for further research.