(Georgia Institute of Technology, 2005)
Su, Jie; Rehg, James M.; Bobick, Aaron F.
Robot navigation using vision is a classic example of a scene understanding problem. We describe a novel approach to estimating the traversability of an unknown environment based on modern object recognition methods. Traversability is an example of an affordance jointly determined by the environment and the physical characteristics of a robot vehicle, whose definition is clear in context. However, it is extremely difficult to estimate the traversability of a given terrain structure in general, or to find rules which work for a wide variety of terrain types. However, by learning to recognize similar terrain structures, it is possible to leverage a limited amount of interaction between the robot and its environment into global statements about the traversability of the scene. We describe a novel on-line learning algorithm that learns to recognize terrain features from images and aggregate the traversability information acquired by a navigating robot. An important property of our method, which is desirable for any learning-based approach to object recognition, is the ability to autonomously acquire arbitrary amounts of training data as needed without any human intervention. Tests of our algorithm on a real robot in complicated unknown natural environments suggest that it is both robust and efficient.
(Georgia Institute of Technology, 2005)
Bobick, Aaron F.; Yang, Zhonghao
In this paper we present the design details of the visual perception system in Aware Home. This paper is intend to (a) detail our design details; (b) provide insights on how different parts of the tracking system are inter-related to solve complex perception tasks; (c) document ideas and potential research directions; (d) provide performance evaluation. This work will primarily focus on algorithm and design issues, while another technical report will be authored to address coding issues related to this system (APIs, class structures, etc).
(Georgia Institute of Technology, 2005)
Bobick, Aaron F.; Essa, Irfan; Shi, Yifan
In this paper, we provide a Bayesian perspective of boosting framework, which we refer to as Bayesian Integration. Through this perspective, we prove the standard ADABOOST is a special case of the naive
Bayesian tree with a mapped conditional probability table and a particular weighting schema. Based on this perspective, we introduce a new algorithm ADABOOST.BAYES by taking the dependency between the
weak classifiers into account, which extends the boosting framework into non-linear combinations of weak classifiers. Compared with standard ADABOOST, ADABOOST.BAYES requires less training iterations but exhibits stronger tendency to overfit. To leverage on both ADABOOST and ADABOOST. BAYES, we introduce a simple switching schema ADABOOST. SOFTBAYES to integrate ADABOOST and ADABOOST.BAYES. Experiments on synthetic data and the UCI data set prove the validity of our framework.