Keywords
Biological motion recognition, Two-Thirds Power Law, Optical Flow, Machine Learning
Abstract
In this work we deal with the problem of designing and developing computational vision models – comparable to the early stages of the human development – using coarse low-level information.
More specifically, we consider a binary classification setting to characterize biological movements with respect to non-biological dynamic events. To this purpose, our model builds on top of the optical flow estimation, and abstract the representation to simulate the limited amount of visual information available at birth. We take inspiration from known biological motion regularities explained by the Two-Thirds Power Law, and design a motion representation that includes different low-level features, which can be interpreted as the computational counterpart of the elements involved in the law.
Our reference application is human-machine interaction, thus the experimental analysis is conducted on a set of videos depicting two different subjects performing a repertoire of dynamic gestures typical of such a setting (e.g. lifting an object, pointing, ...). Two slightly different viewpoints are considered.
The contribution of our work is twofold. First, we show that the effects of the Two-Thirds Power Law can be appreciates on a video analysis setting. Second, we prove that, although the coarse motion representation, our model allows us to reach biological motion classification performances (around 89%) which are reminiscent of the abilities of very young babies. Moreover, our model shows tolerance to view-point changes.
Start Date
13-5-2015 9:55 AM
End Date
13-5-2015 10:20 AM
Session Number
01
Session Title
Motion, Attention, and Eye Movements
Modeling Visual Features to Recognize Biological Motion: A Developmental Approach
In this work we deal with the problem of designing and developing computational vision models – comparable to the early stages of the human development – using coarse low-level information.
More specifically, we consider a binary classification setting to characterize biological movements with respect to non-biological dynamic events. To this purpose, our model builds on top of the optical flow estimation, and abstract the representation to simulate the limited amount of visual information available at birth. We take inspiration from known biological motion regularities explained by the Two-Thirds Power Law, and design a motion representation that includes different low-level features, which can be interpreted as the computational counterpart of the elements involved in the law.
Our reference application is human-machine interaction, thus the experimental analysis is conducted on a set of videos depicting two different subjects performing a repertoire of dynamic gestures typical of such a setting (e.g. lifting an object, pointing, ...). Two slightly different viewpoints are considered.
The contribution of our work is twofold. First, we show that the effects of the Two-Thirds Power Law can be appreciates on a video analysis setting. Second, we prove that, although the coarse motion representation, our model allows us to reach biological motion classification performances (around 89%) which are reminiscent of the abilities of very young babies. Moreover, our model shows tolerance to view-point changes.