The computer vision based applications market is increasing since several years ago. A lot of new... more The computer vision based applications market is increasing since several years ago. A lot of new application areas are incorporating this type of systems to solve particular problems. An emergent trend in last decades is incorporate dedicated devices to particular problems. A smart camera can be defined as a standalone device which can take decisions by itself, since have the ability of understand images. It is composed by an image sensor and processing logic to perform the task, without the use of PCs. The smart cameras can be found, for example, in surveillance, automotive and industry. The book Smart Cameras covers all aspects for these devices from this history and applications, to market trends. The book is composed by 20 chapters organized in eight parts detailed below.Facultad de Informátic
Un control parasitario eficiente permite reducir pérdidas significativas en la agroindustria. Los... more Un control parasitario eficiente permite reducir pérdidas significativas en la agroindustria. Los métodos actuales con los que se realiza este tipo de controles imponen costos y demoras. Por ello, se propone el desarrollo de un dispositivo portátil que automatice esta tarea. En este trabajo se presenta la implementación hardware de un algoritmo de conteo automático de huevos de parásitos utilizando síntesis de alto nivel. Los resultados demuestran la factibilidad de la implementación, con un 87% de precisión operando a una tasa de hasta de 65 frames por segundo y una ocupación de LUTs menor al 45%, considerando dos kits comerciales (PYNQ-Z1 y ULTRA96V2).
Los sistemas de control basado en lógica difusa (FLC, Fuzzy Logic Controller) poseen ventajas, ya... more Los sistemas de control basado en lógica difusa (FLC, Fuzzy Logic Controller) poseen ventajas, ya que no requieren modelado matemático y además son útiles cuando se necesita del conocimiento de un experto para el manejo de datos imprecisos. Resulta interesante para la implementación de un FLCs la utilización de la tecnología FPGA. Esta tecnología presenta ventajas respecto a la velocidad de procesamiento, consumo de potencia, flexibilidad de diseño y reconfiguración. Este trabajo presenta una herramienta basada en HLS para generar FLCs sobre FPGA. Durante su desarrollo también se analizaron las directivas de síntesis con mayor impacto sobre la performance de los algoritmos. Además se desarrollaron tres problemas de lógica difusa para verificar el funcionamiento de la herramienta.
[10 p.]Las Máquinas de Soporte Vectorial (SVM) son una herramienta de aprendizaje común utilizada... more [10 p.]Las Máquinas de Soporte Vectorial (SVM) son una herramienta de aprendizaje común utilizada ampliamente gracias a su alta precisión en tareas de clasificación. El uso de las SVM en tiempo real a través de sistemas embebidos resulta desafiante debido a la complejidad de los cálculos que se requieren. Por tanto, se hace presente la necesidad de migrar las SVM a plataformas hardware con el fin de alcanzar un alto rendimiento con costes y consumos de energía bajos. El presente artículo proporciona un estudio exhaustivo del proceso de aceleración de la librería LIBSVM mediante hardware. Dicha librería será utilizada durante la clasificación binaria de características, tomando como caso de estudio la identificación de masas mamarias. Se presenta además una optimización que ofrece una aceleración del 57% respecto de algoritmo original. Por último, se sugieren algunas direcciones claves que optimizarán el proceso de aceleración en futuras investigaciones.Fil: Todorovich, Elías. Universidad Católica de Salta. Facultad de Ingeniería e Informática; Argentina.Fil: Puig, Domènec. Universidad Católica de Salta. Facultad de Ingeniería e Informática; Argentina.Fil: Vázquez, Martín. Universidad Católica de Salta. Facultad de Ingeniería e Informática; Argentina.Fil: Torrents-Barrena, Jordina. Universidad Católica de Salta. Facultad de Ingeniería e Informática; Argentina.Fil: Leiva, Lucas. Universidad Católica de Salta. Facultad de Ingeniería e Informática; Argentina
La modernización en las fábricas es un factor clave para la producción y la calidad del producto ... more La modernización en las fábricas es un factor clave para la producción y la calidad del producto final. Sin embargo, esta modernización puede suponer una inversión que las empresas no pueden asumir, dejándolas fuera de la adaptación a la Industria 4.0. En la industria de fabricación de baldosas cerámicas se utilizan inspecciones visuales para determinar la calidad del producto final. Estas tareas son realizadas generalmente por operarios expuestos a entornos de riesgo. Este trabajo presenta una solución de bajo costo para la inspección automática de baldosas cerámicas. Los defectos analizados son gotas, defectos de material, esquinas, bordes y dimensiones. Todos los algoritmos se implementaron en SoC FPGA (dispositivo Xilinx Zynq) utilizando síntesis de alto nivel. Los algoritmos se verificaron y validaron en un entorno controlado construido para evaluar aplicaciones de inspección visual. Los resultados de utilización de recursos y tiempos de procesamiento indican que la implementa...
The computer vision based applications market is increasing since several years ago. A lot of new... more The computer vision based applications market is increasing since several years ago. A lot of new application areas are incorporating this type of systems to solve particular problems. An emergent trend in last decades is incorporate dedicated devices to particular problems. A smart camera can be defined as a standalone device which can take decisions by itself, since have the ability of understand images. It is composed by an image sensor and processing logic to perform the task, without the use of PCs. The smart cameras can be found, for example, in surveillance, automotive and industry. The book Smart Cameras covers all aspects for these devices from this history and applications, to market trends. The book is composed by 20 chapters organized in eight parts detailed below.
Vision based applications are present anywhere. A special market is industry, allowing to improve... more Vision based applications are present anywhere. A special market is industry, allowing to improve product quality and to reduce manufacturing costs. The vision systems applied to industries are known as machine vision systems. These systems must meet time constraints to operate in real time. Generally the production lines are more and more fasters, and the time to process and bring a response is minimal. For this reasons, dedicated architectures are emplaced. In this work a review of several commercial systems is presented, as well a proposed architecture is depicted. The architecture is concern as a customizable platform, avoiding having knowledge in hardware description languages. It is based on massive parallelism to achieve the maximum processing performance. Several optimizations at different levels are applied to increase the final system speedup. Also, time and area metrics are reported, showing that the architecture is well suitable for real time video processing in industri...
Abstract. This paper presents a parallel architecture for a radial basis function (RBF) neural ne... more Abstract. This paper presents a parallel architecture for a radial basis function (RBF) neural network used for pattern recognition. This architecture allows defining sub-networks which can be activated sequentially. It can be used as a fruitful classification mechanism in many application fields. Several implementations of the network on a Xilinx FPGA Virtex 4- (xc4vsx25) are presented, with speed and area evaluation metrics. Some network improvements have been achieved by segmenting the critical path. The results expressed in terms of speed and area are satisfactory and have been applied to pattern recognition problems.
Vision based applications are present anywhere. A special market is industry, allowing to improve... more Vision based applications are present anywhere. A special market is industry, allowing to improve product quality and to reduce manufacturing costs. The vision systems applied to industries are known as machine vision systems. These systems must meet time constraints to operate in real time. Generally the production lines are more and more fasters, and the time to process and bring a response is minimal. For this reasons, dedicated architectures are emplaced. In this work a review of several commercial systems is presented, as well a proposed architecture is depicted. The architecture is concern as a customizable platform, avoiding having knowledge in hardware description languages. It is based on massive parallelism to achieve the maximum processing performance. Several optimizations at different levels are applied to increase the final system speedup. Also, time and area metrics are reported, showing that the architecture is well suitable for real time video processing in industri...
In this paper a compiler capable of generate Multiple Instruction Single Data (MISD) architecture... more In this paper a compiler capable of generate Multiple Instruction Single Data (MISD) architectures for feature vector calculation is presented. The input is a high-level language, avoiding to developers to involve in low level design. Instead, the output is expressed in a Hardware Description Language (HDL), and can be used for FPGA configuration. A FPGA is a programmable device which allows parallelism, increasing the system speed up. The tool is intended to use in feature vector calculation of region of interest (ROI) for real time video applications. These ROIs arrives serially. Also, is possible to evaluate the vector in design time, allowing system prototyping. The compiler optimizes the response time and the number of registers required to meet real time constraints.
The use of image processing systems is becoming frequent and is appropriate in edge computing. Ho... more The use of image processing systems is becoming frequent and is appropriate in edge computing. However, the demands on power consumption and high performance prevent the use of standard processing platforms. Thus, FPGAs are a good option for the development of computer vision systems because they are capable of exploiting parallelism. On the other hand, the design flow of current FPGA synthesis tools supports high-level languages as input descriptions, in contrast to hardware description languages. High-level synthesis (HLS) automates the design process by transforming an algorithmic description into digital hardware meeting design limitations. However, image processing experts may find challenging the hardware integration with the rest of the system components, e.g. capture and display interfaces. This work presents a basic design for the construction of image processing applications based on Zynq and a guide for its use, which solves this problem, allowing the agile generation of ...
The computer vision based applications market is increasing since several years ago. A lot of new... more The computer vision based applications market is increasing since several years ago. A lot of new application areas are incorporating this type of systems to solve particular problems. An emergent trend in last decades is incorporate dedicated devices to particular problems. A smart camera can be defined as a standalone device which can take decisions by itself, since have the ability of understand images. It is composed by an image sensor and processing logic to perform the task, without the use of PCs. The smart cameras can be found, for example, in surveillance, automotive and industry. The book Smart Cameras covers all aspects for these devices from this history and applications, to market trends. The book is composed by 20 chapters organized in eight parts detailed below.Facultad de Informátic
Un control parasitario eficiente permite reducir pérdidas significativas en la agroindustria. Los... more Un control parasitario eficiente permite reducir pérdidas significativas en la agroindustria. Los métodos actuales con los que se realiza este tipo de controles imponen costos y demoras. Por ello, se propone el desarrollo de un dispositivo portátil que automatice esta tarea. En este trabajo se presenta la implementación hardware de un algoritmo de conteo automático de huevos de parásitos utilizando síntesis de alto nivel. Los resultados demuestran la factibilidad de la implementación, con un 87% de precisión operando a una tasa de hasta de 65 frames por segundo y una ocupación de LUTs menor al 45%, considerando dos kits comerciales (PYNQ-Z1 y ULTRA96V2).
Los sistemas de control basado en lógica difusa (FLC, Fuzzy Logic Controller) poseen ventajas, ya... more Los sistemas de control basado en lógica difusa (FLC, Fuzzy Logic Controller) poseen ventajas, ya que no requieren modelado matemático y además son útiles cuando se necesita del conocimiento de un experto para el manejo de datos imprecisos. Resulta interesante para la implementación de un FLCs la utilización de la tecnología FPGA. Esta tecnología presenta ventajas respecto a la velocidad de procesamiento, consumo de potencia, flexibilidad de diseño y reconfiguración. Este trabajo presenta una herramienta basada en HLS para generar FLCs sobre FPGA. Durante su desarrollo también se analizaron las directivas de síntesis con mayor impacto sobre la performance de los algoritmos. Además se desarrollaron tres problemas de lógica difusa para verificar el funcionamiento de la herramienta.
[10 p.]Las Máquinas de Soporte Vectorial (SVM) son una herramienta de aprendizaje común utilizada... more [10 p.]Las Máquinas de Soporte Vectorial (SVM) son una herramienta de aprendizaje común utilizada ampliamente gracias a su alta precisión en tareas de clasificación. El uso de las SVM en tiempo real a través de sistemas embebidos resulta desafiante debido a la complejidad de los cálculos que se requieren. Por tanto, se hace presente la necesidad de migrar las SVM a plataformas hardware con el fin de alcanzar un alto rendimiento con costes y consumos de energía bajos. El presente artículo proporciona un estudio exhaustivo del proceso de aceleración de la librería LIBSVM mediante hardware. Dicha librería será utilizada durante la clasificación binaria de características, tomando como caso de estudio la identificación de masas mamarias. Se presenta además una optimización que ofrece una aceleración del 57% respecto de algoritmo original. Por último, se sugieren algunas direcciones claves que optimizarán el proceso de aceleración en futuras investigaciones.Fil: Todorovich, Elías. Universidad Católica de Salta. Facultad de Ingeniería e Informática; Argentina.Fil: Puig, Domènec. Universidad Católica de Salta. Facultad de Ingeniería e Informática; Argentina.Fil: Vázquez, Martín. Universidad Católica de Salta. Facultad de Ingeniería e Informática; Argentina.Fil: Torrents-Barrena, Jordina. Universidad Católica de Salta. Facultad de Ingeniería e Informática; Argentina.Fil: Leiva, Lucas. Universidad Católica de Salta. Facultad de Ingeniería e Informática; Argentina
La modernización en las fábricas es un factor clave para la producción y la calidad del producto ... more La modernización en las fábricas es un factor clave para la producción y la calidad del producto final. Sin embargo, esta modernización puede suponer una inversión que las empresas no pueden asumir, dejándolas fuera de la adaptación a la Industria 4.0. En la industria de fabricación de baldosas cerámicas se utilizan inspecciones visuales para determinar la calidad del producto final. Estas tareas son realizadas generalmente por operarios expuestos a entornos de riesgo. Este trabajo presenta una solución de bajo costo para la inspección automática de baldosas cerámicas. Los defectos analizados son gotas, defectos de material, esquinas, bordes y dimensiones. Todos los algoritmos se implementaron en SoC FPGA (dispositivo Xilinx Zynq) utilizando síntesis de alto nivel. Los algoritmos se verificaron y validaron en un entorno controlado construido para evaluar aplicaciones de inspección visual. Los resultados de utilización de recursos y tiempos de procesamiento indican que la implementa...
The computer vision based applications market is increasing since several years ago. A lot of new... more The computer vision based applications market is increasing since several years ago. A lot of new application areas are incorporating this type of systems to solve particular problems. An emergent trend in last decades is incorporate dedicated devices to particular problems. A smart camera can be defined as a standalone device which can take decisions by itself, since have the ability of understand images. It is composed by an image sensor and processing logic to perform the task, without the use of PCs. The smart cameras can be found, for example, in surveillance, automotive and industry. The book Smart Cameras covers all aspects for these devices from this history and applications, to market trends. The book is composed by 20 chapters organized in eight parts detailed below.
Vision based applications are present anywhere. A special market is industry, allowing to improve... more Vision based applications are present anywhere. A special market is industry, allowing to improve product quality and to reduce manufacturing costs. The vision systems applied to industries are known as machine vision systems. These systems must meet time constraints to operate in real time. Generally the production lines are more and more fasters, and the time to process and bring a response is minimal. For this reasons, dedicated architectures are emplaced. In this work a review of several commercial systems is presented, as well a proposed architecture is depicted. The architecture is concern as a customizable platform, avoiding having knowledge in hardware description languages. It is based on massive parallelism to achieve the maximum processing performance. Several optimizations at different levels are applied to increase the final system speedup. Also, time and area metrics are reported, showing that the architecture is well suitable for real time video processing in industri...
Abstract. This paper presents a parallel architecture for a radial basis function (RBF) neural ne... more Abstract. This paper presents a parallel architecture for a radial basis function (RBF) neural network used for pattern recognition. This architecture allows defining sub-networks which can be activated sequentially. It can be used as a fruitful classification mechanism in many application fields. Several implementations of the network on a Xilinx FPGA Virtex 4- (xc4vsx25) are presented, with speed and area evaluation metrics. Some network improvements have been achieved by segmenting the critical path. The results expressed in terms of speed and area are satisfactory and have been applied to pattern recognition problems.
Vision based applications are present anywhere. A special market is industry, allowing to improve... more Vision based applications are present anywhere. A special market is industry, allowing to improve product quality and to reduce manufacturing costs. The vision systems applied to industries are known as machine vision systems. These systems must meet time constraints to operate in real time. Generally the production lines are more and more fasters, and the time to process and bring a response is minimal. For this reasons, dedicated architectures are emplaced. In this work a review of several commercial systems is presented, as well a proposed architecture is depicted. The architecture is concern as a customizable platform, avoiding having knowledge in hardware description languages. It is based on massive parallelism to achieve the maximum processing performance. Several optimizations at different levels are applied to increase the final system speedup. Also, time and area metrics are reported, showing that the architecture is well suitable for real time video processing in industri...
In this paper a compiler capable of generate Multiple Instruction Single Data (MISD) architecture... more In this paper a compiler capable of generate Multiple Instruction Single Data (MISD) architectures for feature vector calculation is presented. The input is a high-level language, avoiding to developers to involve in low level design. Instead, the output is expressed in a Hardware Description Language (HDL), and can be used for FPGA configuration. A FPGA is a programmable device which allows parallelism, increasing the system speed up. The tool is intended to use in feature vector calculation of region of interest (ROI) for real time video applications. These ROIs arrives serially. Also, is possible to evaluate the vector in design time, allowing system prototyping. The compiler optimizes the response time and the number of registers required to meet real time constraints.
The use of image processing systems is becoming frequent and is appropriate in edge computing. Ho... more The use of image processing systems is becoming frequent and is appropriate in edge computing. However, the demands on power consumption and high performance prevent the use of standard processing platforms. Thus, FPGAs are a good option for the development of computer vision systems because they are capable of exploiting parallelism. On the other hand, the design flow of current FPGA synthesis tools supports high-level languages as input descriptions, in contrast to hardware description languages. High-level synthesis (HLS) automates the design process by transforming an algorithmic description into digital hardware meeting design limitations. However, image processing experts may find challenging the hardware integration with the rest of the system components, e.g. capture and display interfaces. This work presents a basic design for the construction of image processing applications based on Zynq and a guide for its use, which solves this problem, allowing the agile generation of ...
Uploads
Papers by Lucas Leiva