Workshop: Xuan Huang and Peter Sandilands

Xuan Huang

Fish recognition:  A hierarchical approach

Live fish recognition in the open sea is a challenging multi-class classification task.  We propose a hierarchical classification approach to recognize live fish from underwater videos. From the detected fish images, 66 types of feature are extracted.  These features are a combination of color, shape and texture properties in different parts of the fish such as tail/head/top/bottom, as well as the whole fish. We create a taxonomy knowledge based tree and compare it to state-of-the-art techniques on a live fish image dataset.

Peter Sandilands

Generalised Hand-Object Interactions

Interactions with objects are very important in virtual worlds. A large part of the experience in these worlds comes from being able to affect the environment, and do it in a plausible manner. Current methods of retargeting motion to the environment either require a manual pre-processing stage, do not maintain the naturalness of the original motion, or both. We hypothesise that by an improved representation for motion in these types of scenes, creating original, natural-looking interactions in real-time is possible.

In this workshop, I will describe our approach to addressing this problem, showing some entirely automatic, faster than real-time grasping results, a low-cost motion-capture technique which takes into account the interaction with objects, and an 'Interaction mesh'-inspired technique for retargeting close interactions with objects.