| Automated visual recognition and detection processes are becoming increasingly prevalent in almost all scientific fields and being currently implemented in many fields of industry. In most cases, systems are painstakingly designed and developed in order to detect only a single and specific object or property of an object. The objective of this project was to create a framework of development in which any object distinguishable in a two-dimensional digital image could be analyzed and subsequently detected in other images. Furthermore, as new methods are developed, they could be easily incorporated into this framework to ultimately improve the performance of the system.; This thesis describes a highly adaptable, general-application visual detection system as well as several innovative methods for the description of objects without which such adaptivity would be impossible. Two-dimensional, still images are analyzed and objects of interest can be introduced to the system. Objects are then described by a variety of properties through derived attributes and stored in a database. Occurrences of these objects can then be detected in future images through comparisons to selected models. The system is fully expandable in that new properties and comparison techniques or criteria can be added as they are developed and as their need becomes apparent. The system is presented with a basic set of attribute representations and methods of comparison, and their development and origin are described in detail. The database structure is outlined and the process by which new properties and comparative methods can be added is described. Seventeen different images containing nearly two thousand separate objects were searched for various model objects and the average classification accuracy was 98.3%. In most images, more than 100 object classifications could be performed per second at an accuracy higher than 95% when no higher order analyses were required. |
