Implementation of models for image processing applications with real-time constraints

Resumen

This work has diverse research branchs, one of them is the implementation for high performance computing, while the other branchs are focused on computer models and bio-inspired algorithms. In this thesis, it has developed different computer vision models for application in automotive, biomedical applications and food industry. Therefore, the main contribution of this work is the applied component, which requires an efficient implementation and processing in specific purpose architectures (for embedded systems) or high performance (to provide results and simulations efficiently, eg shortening time in diagnosis and monitoring of diseases). This thesis is divided into five sections: In the first and second section consists in a brief introduction (English and Spanish respectively) to computer vision where it is explained the typical steps that are performed in this area to solve an especific problems. The third section proposes an adaptation of an efficient optical flow model in a hybrid architecture (focal plane and soft-processor), with the objective of developing a real time system that work as driver assistant and advises to the driver about vehicles in overtaking to avoid unexpected maneuvers that could cause an accident. The forth section of this thesis is focused on the field of ultrasound, where it will be exposed a method based on an evolutive algorithm for segmenting the carotid artery to obtain an accurate area estimation and determinate the degree of stenosis of the patient. Additionally it is proposed a model to estimate sub-pixel motion, in combination with other techniques, to measure the wall artery displacement in order to determinate possible pathologies of the subject. The fifth section of this thesis is focused to design an experimental system in real time to analyze images in field of food industry. This system is able to analyze and process images, in order to reject those products that are defective. The last section summarizes the final conclusions of this thesis and exposed future works.