Browsing by Issue Date, starting with "2012-02-22"
Now showing 1 - 1 of 1
Results Per Page
Sort Options
- Fuzzy logic based approach for object feature trackingPublication . Lopes, Nuno Vieira; Melo-Pinto, Pedro; Couto, Pedro M.This thesis introduces a novel technique for feature tracking in sequences of greyscale images based on fuzzy logic. A versatile and modular methodology for feature tracking using fuzzy sets and inference engines is presented. Moreover, an extension of this methodology to perform the correct tracking of multiple features is also presented. To perform feature tracking three membership functions are initially defined. A membership function related to the distinctive property of the feature to be tracked. A membership function is related to the fact of considering that the feature has smooth movement between each image sequence and a membership function concerns its expected future location. Applying these functions to the image pixels, the corresponding fuzzy sets are obtained and then mathematically manipulated to serve as input to an inference engine. Situations such as occlusion or detection failure of features are overcome using estimated positions calculated using a motion model and a state vector of the feature. This methodology was previously applied to track a single feature identified by the user. Several performance tests were conducted on sequences of both synthetic and real images. Experimental results are presented, analysed and discussed. Although this methodology could be applied directly to multiple feature tracking, an extension of this methodology has been developed within that purpose. In this new method, the processing sequence of each feature is dynamic and hierarchical. Dynamic because this sequence can change over time and hierarchical because features with higher priority will be processed first. Thus, the process gives preference to features whose location are easier to predict compared with features whose knowledge of their behavior is less predictable. When this priority value becomes too low, the feature will no longer tracked by the algorithm. To access the performance of this new approach, sequences of images where several features specified by the user are to be tracked were used. In the final part of this work, conclusions drawn from this work as well as the definition of some guidelines for future research are presented.