Object recognition in videos is one of the main challenges in computer vision. Several methods have been proposed to achieve this task, such as background subtraction, temporal differencing, optical flow, particle filtering among others. Since the introduction of Convolutonal Neural Networks (CNN) for object detection in the Imagenet Large Scale Visual Recognition Competition (ILSVRC), its use for image detection and classification has increased, becoming the state-of-the-art for such task, being Faster R-CNN the preferred model in the latest ILSVRC challenges. Moreover, the Faster R-CNN model, with minimum modifications, has been succesfully used to detect and classify objects (either static or dynamic) in video sequences; in such setup, the frames of the video are input “as is” i.e. without any pre-processing. In this thesis work we propose to use Robust PCA (RPCA, a.k.a. Principal Component Pursuit, PCP), as a video background modeling pre-processing step, before using the Faster R-CNN model, in order to improve the overall performance of detection and classification of, specifically, the moving objects. We hypothesize that such pre-processing step, which segments the moving objects from the background, would reduce the amount of regions to be analyzed in a given frame and thus (i) improve the classification time and (ii) reduce the error in classification for the dynamic objects present in the video. In particular, we use a fully incremental RPCA / PCP algorithm that is suitable for real-time or on-line processing. Furthermore, we present extensive computational results that were carried out in three different platforms: A high-end server with a Tesla K40m GPU, a desktop with a Tesla K10m GPU and the embedded system Jetson TK1. Our classification results attain competitive or superior performance in terms of Fmeasure, achieving an improvement ranging from 3.7% to 97.2%, with a mean improvement of 22% when the sparse image was used to detect and classify the object with the neural network, while at the same time, reducing the classification time in all architectures by a factor raging between 2% and 25%. / Tesis

Object recognition in videos is one of the main challenges in computer vision. Several methods have been proposed to achieve this task, such as background subtraction, temporal differencing, optical flow, particle filtering among others. Since the introduction of Convolutonal Neural Networks (CNN) for object detection in the Imagenet Large Scale Visual Recognition Competition (ILSVRC), its use for image detection and classification has increased, becoming the state-of-the-art for such task, being Faster R-CNN the preferred model in the latest ILSVRC challenges. Moreover, the Faster R-CNN model, with minimum modifications, has been succesfully used to detect and classify objects (either static or dynamic) in video sequences; in such setup, the frames of the video are input “as is” i.e. without any pre-processing. In this thesis work we propose to use Robust PCA (RPCA, a.k.a. Principal Component Pursuit, PCP), as a video background modeling pre-processing step, before using the Faster R-CNN model, in order to improve the overall performance of detection and classification of, specifically, the moving objects. We hypothesize that such pre-processing step, which segments the moving objects from the background, would reduce the amount of regions to be analyzed in a given frame and thus (i) improve the classification time and (ii) reduce the error in classification for the dynamic objects present in the video. In particular, we use a fully incremental RPCA / PCP algorithm that is suitable for real-time or on-line processing. Furthermore, we present extensive computational results that were carried out in three different platforms: A high-end server with a Tesla K40m GPU, a desktop with a Tesla K10m GPU and the embedded system Jetson TK1. Our classification results attain competitive or superior performance in terms of Fmeasure, achieving an improvement ranging from 3.7% to 97.2%, with a mean improvement of 22% when the sparse image was used to detect and classify the object with the neural network, while at the same time, reducing the classification time in all architectures by a factor raging between 2% and 25%. / Tesis

En la presente tesis, queremos demostrar y proponer como la tecnología puede ser utilizada por los genetistas y especialistas en oncología como una herramienta para agilizar la detección de cáncer de mama, siendo este el más común en Perú. El diagnóstico temprano es un mecanismo efectivo que ayuda a la reducción de la mortalidad en este tipo de cáncer de tal manera que se pueda seguir un tratamiento adecuado. Actualmente una forma de detectarlo es a través de una prueba genética para identificar mutaciones en los genes BRCA 1 y BRCA 2, sin embargo, este camino contiene pruebas que son difíciles, costosas y lentas, que a su vez requieren una carga de trabajo excesiva por parte de un biólogo o genetista. por tal motivo se tiene como objetivo combinar los factores de riesgo asociados con el cáncer de mamá, incluidas las variaciones genéticas para diseñar un modelo predictivo basados en la inteligencia artificial para determinar si el tumor asociado al cáncer es benigno o maligno. El modelo se diseñó utilizando un algoritmo de redes neuronales logrando obtener un rendimiento de 92% precisión con datos de prueba en tan solo unos minutos. Esta propuesta de modelo de predicción es única en el Perú y puede ser ofrecida por una Gerencia de TI dentro de una organización del sector salud para que posteriormente pueda ser implementada y desplegada por un equipo de científicos de datos. / In the present thesis, we are looking for a demonstration and proposal how the technology can be so useful for the genetic and oncology Scientifics as a tool for quick detection of the breast cancer, which ones is the most common in Peru. Early diagnosis is the most effective way for a treatment to help people to prevent the mortality in this kind of cancer. At this moment, the best way for an early detection is a genetical test to look for mutations in BRCA 1 and BRCA 2 gen, however this way is so hard, because this requires a lot of difficult, expensive, and slowly tests remark a lot of work of the genetic and oncology Scientifics. That is the reason our thesis has as the principal goal to combine all the risk factors associated with breast cancer, including genetical mutations, for generate a predictive model based in artificial intelligence for determinate if a kind of tumor is associated with benign or pathogenic. This designed model has a 92% of precision with open-source test data in a few minutes. This predictive model is unique in Peru and can be offered by an IT Management within a health sector organization so that it can later be implemented and deployed by a team of data scientists.