Natural Object Recognition. Questo articolo non è disponibile.Natural Object Recognition

Natural Object Recognition presents a totally new approach to the automation of scene understanding. Rather than attempting to construct highly specialized algorithms for recognizing physical objects, as is customary in modern computer vision research, the application and subsequent evaluation of large numbers of relatively straightforward image processing routines is used to recognize natural features such as trees, bushes, and rocks. The use of contextual information is the key to simplifying the problem to the extent that well understood algorithms give reliable results in ground-level, outdoor scenes.

1 Introduction.- 1.1 Motivation.- 1.2 Issues.- 1.3 Contribution.- 1.4 Example of results.- 2 Natural Object Recognition.- 2.1 Visual capabilities for autonomous robots.- 2.2 Related research.- 2.2.1 Recognizing objects.- 2.2.2 Recognizing natural scenes.- 2.3 Limitations of current machine-vision technology.- 2.3.1 Shape.- 2.3.2 Universal partitioning.- 2.3.3 Contextual knowledge.- 2.3.4 Computational complexity.- 2.4 Key ideas.- 2.4.1 Context-limited vision.- 2.4.2 Global consistency.- 2.4.3 Candidate comparison to control complexity.- 2.4.4 Layered partitions.- 2.5 Experimental results.- 2.6 Conclusions.- 3 A Vision System for off-Road Navigation.- 3.1 Task scenario.- 3.2 Prior knowledge.- 3.3 The role of geometry.- 3.3.1 Sources and limitations of range data.- 3.3.2 Using three-dimensional geometric information.- 3.4 A vocabulary for recognition.- 3.4.1 Target vocabulary.- 3.4.2 Recognition vocabulary.- 3.5 Contextual information.- 3.5.1 Types of context.- 3.5.2 Using context.- 4 Context-Based Vision.- 4.1 Conceptual Architecture.- 4.1.1 Overview.- 4.1.2 Context sets.- 4.1.3 Candidate generation.- 4.1.4 Clique formation.- 4.1.5 Candidate comparison.- 4.1.6 The recognition process.- 4.2 Implementation of Condor.- 4.2.1 Processing sequence.- 4.2.2 Representation of context.- 4.2.3 Context-set construction.- 4.3 Example of natural-object recognition.- 4.3.1 Candidate generation.- 4.3.2 Candidate comparison.- 4.3.3 Clique formation.- 4.3.4 Clique selection.- 4.4 Automated knowledge acquisition.- 4.4.1 Learning a description of the environment.- 4.4.2 Applying recognition procedures more effectively.- 4.4.3 Refining context sets.- 4.4.4 Discovering new procedures.- 4.5 Complexity analysis.- 4.5.1 Mathematical details.- 4.5.2 The number of cliques.- 4.6 Discussion.- 4.6.1 Reliability.- 4.6.2 Recognition strategies.- 4.6.3 Knowledge-base construction.- 4.6.4 Multiple-image interpretation.- 4 Context-Based Vision.- 5.1 Evaluation scenario.- 5.2 Experimentation.- 5.2.1 Experiment 1.- 5.2.2 Experiment 2.- 5.2.3 Experiment 3.- 5.3 Analysis of results.- 5.3.1 Fixing mistakes.- 5.3.2 Accounting for success.- 5.3.3 Evaluating relevance to the goal.- 6 Conclusion.- 6.1 Contribution.- 6.2 Evaluation.- 6.2.1 Competence.- 6.2.2 Scalability.- 6.2.3 Generality.- 6.2.4 Evaluation plan.- 6.3 Summary.- A The Core Knowledge Structure.- A.1 Introduction.- A.2 Core Knowledge Structure.- A.3 Logical Interpretation of the CKS Database.- A.3.1 Semantics.- A.3.2 Insertions.- A.3.3 Queries.- A.3.4 User-Defined Relations.- A.3.5 Discussion.- A.4 Slot Access.- A.5 Summary.- References.