Ciencias Exactas y Ciencias de la Salud
Permanent URI for this collectionhttps://hdl.handle.net/11285/551039
Pertenecen a esta colección Tesis y Trabajos de grado de las Maestrías correspondientes a las Escuelas de Ingeniería y Ciencias así como a Medicina y Ciencias de la Salud.
Browse
Search Results
- End-to-End Violence Detection Using Pedestrian Detection, Pose Estimation, and Temporal GRUs for Surveillance Applications(Instituto Tecnológico y de Estudios Superiores de Monterrey, 2025-05-27) Salazar Vasquez, Fredy Antonio; Conant Pablos, Santiago Enrique; emipsanchez; Ortiz bayliss, José Carlos; School of Engineering and Sciences; Campus MonterreyIn recent years, surveillance systems have played an increasingly prominent role in both public and private settings. These systems monitor activities in real time and provide information to security personnel and authorities. Their constant observation helps prevent incidents and maintain order. Traditional surveillance systems record events but do not fully exploit the valuable information they capture. New technologies allow valuable data to be extracted, turning surveillance into an active tool for security. With the development of tools like object detection, pose estimation, and neural networks, surveillance systems can now interpret the scenes they capture. Rather than simply recording footage, these systems are becoming active participants in security by extracting meaningful information from visual data. Despite these advances, it remains a challenge to identify violent acts using visual information. The main challenge is to analyze the data in a way that identifies risks. Although cameras capture lot of information, traditional systems do not always use them preventively. These systems must predict risky situations by detecting aggressive behavior or suspicious activities early. This work primarily focuses on addressing the development of techniques to improve the detection of violence in surveillance videos by optimizing specific processes such as pedestrian detection, human posture estimation, object tracking, and violent behavior classification. Pedestrian detection is optimized using advanced models like YOLO, enhancing accuracy in high-density environments. Posture estimation is improved through advanced pose detection algorithms that reduce manual intervention. Object tracking is enhanced by implementing Deep SORT to maintain reliable identity tracking across video frames. Violent behavior classification is fine-tuned using a deep neural network architecture based on Gated Recurrent Units (GRU), which captures temporal movement patterns. Video footage from the KranokNV database is processed to identify joint angles of pedestrians, and the VID dataset is used to evaluate system performance. This integrated approach aims to achieve faster, more accurate, and more reliable detection of violent situations, contributing to public safety. Additionally, the evaluation considers spatial and temporal features, such as velocity, acceleration, motion energy, abrupt changes, symmetry, and expansion radius. The processed data was smoothed with the Kalman filter, achieving an accuracy of 99.44%. The results indicate continuous detection capability and improvement in generalization throughout the training process.
- Aspect based sentiment analysis in students’ evaluation of teaching(Instituto Tecnológico y de Estudios Superiores de Monterrey, 2024-05) Acosta Ugalde, Diego; Conant Pablos, Santiago Enrique; mtyahinojosa, emipsanchez; Guitérrez Rodríguez, Andrés Eduardo; Juárez Jiménez, Julio Antonio; Morales Méndez, Rubén; School of Engineering and Sciences; Campus Monterrey; Camacho Zuñiga, ClaudiaStudent evaluations of teachings (SETs) are essential for assessing educational quality. Natural Language Processing (NLP) techniques can produce informative insights from these evaluations. The large quantity of text data received from SETs has surpassed the capacity for manual processing. Employing NLP to analyze student feedback offers an efficient method for understanding educational experiences, enabling educational institutions to identify patterns and trends that might have been difficult, if not impossible, to notice with a manual analysis. Data mining using NLP techniques can delve into the thoughts and perspectives of students on their educational experiences, identifying sentiments and aspects that may have a level of abstraction that the human analysis cannot perceive. I use different NLP techniques to enhance the analysis of student feedback in the form of comments and provide better insights and understanding into factors that influence students’ sentiments. This study aims to provide an overview of the various approaches used in NLP and sentiment analysis, focusing on analyzing the models and text representations used to classify numerical scores obtained from the text feedback of a corpus of SETs in Spanish. I provide a series of experiments using different text classification algorithms for sentiment classification over numerical scores of educational aspects. Additionally, I explore two Aspect Based Sentiment Analysis (ABSA) models, a pipeline and a multi-task approach, to extract broad and comprehensive insights from educational feedback for each professor. The results of this research demonstrate the effectiveness of using NLP techniques for analyzing student feedback. The sentiment classification experiments showed favorable outcomes, indicating that it is possible to utilize student comments to classify certain educational scores accurately. Furthermore, the qualitative results obtained from the ABSA models, presented in a user-friendly dashboard, highlight the efficiency and utility of employing these algorithms for the analysis of student feedback. The dashboard provides valuable insights into the sentiments expressed by students regarding various aspects of their educational experience, allowing for a more comprehensive understanding of the factors influencing their opinions. These findings highlight the potential of NLP in the educational domain, offering a powerful tool for institutions to gain a deeper understanding of student perspectives and make data-driven decisions to enhance the quality of education.
- Exploring data-driven selection hyper-heuristic approaches for the curriculum-based course timetabling(Instituto Tecnológico y de Estudios Superiores de Monterrey, 2020-12) Hinojosa Cavada, Carlos Alfonso; CONANT PABLOS, SANTIAGO ENRIQUE; 56551; Conant Pablos, Santiago Enrique; emipsanchez; Ortiz Bayliss, José Carlos; School of Engineering and Sciences; Campus MonterreyThe curriculum-based timetabling problem (CB-CTT) represents a challenging field of study within educational timetabling, with real-world applications that stress its importance. Solving a CB-CTT problem requires allocating a set of courses using limited resources, subject to a set of hard constraints that must be satisfied. The goal then is to find a feasible assignment for every lecture that constitutes the courses to the positions in the timetable formed by a combination of day, period, and room; all while minimizing an objective function as specified by the constraints in the problem. Designing the timetable for the courses in the incoming term is a problem faced by universities each academic period. Given the complexity of manually designing timetables, automated methods have attracted the attention of many researchers for solving this problem. The design of timetables remains an open problem to this day. According to the no free lunch theorem, different heuristics are effective on different problem instances, stressing the importance of finding automated methods for designing timetables. This dissertation explores novel hyper-heuristic models that rely on various machine learning techniques, such as boosting, clustering and principal component analysis. In total, two models were designed and implemented as results of this work. The first model relies on gradient boosting algorithms to generate a selection hyper-heuristic. The general idea is that different instances of the CB-CTT are best solved by different heuristics. Hence, the aim is to create a model that learns from the features that describe problem instances and predicts which would be the most suitable heuristic to apply. While the classification model produces promising results in terms of accuracy, the quality of the generated solutions is bounded by the best-known single heuristic. The second model aims to remove the bounds set by the use of a single heuristic by exploring ways of combining heuristics during the timetable construction process. The selection hyper-heuristic approach is powered by principal component analysis and k-means. The model starts by identifying similar regions in the instance space and keeping track of the performance of each heuristic for those regions. Then, when constructing new timetables, the model determines the most suitable heuristic for a given region of the instance space. The method was able to outperform the synthetic oracle created by taking the result of the best isolated heuristic in several instances. This dissertation is submitted to the Graduate Programs in Engineering and Information Technologies in partial fulfillment of the requirements for the degree of Master of Science in Computer Sciences with a major in Intelligent Systems.
- Automatic placement of electronic components to maximize heat dissipation on PCB's using particle swarm optimization(2017-12) Ramirez Velazco, Omar Alexander; Conant Pablos, Santiago Enrique; Ortiz Bayliss, José Carlos; Amaya Contreras, Iván MauricioThis thesis documents my personal research as candidate for the academic degree of Master of Science in Intelligent Systems. The purpose of this work is to optimize the electronic components layout in a printed circuit board based on metrics related to its thermal dissipation. The continuous evolution of smaller, more complex and compact integrated circuits pushes the design and fabrication techniques of printed circuit board, also known as PCB, to new limits. Moore’s Law states: “The number of transistors incorporated in a chip will approximately double every 24 months.” Nowadays, it is possible to see several examples of this in the new microprocessors used in computers, smart phones and tablets that have more than 600 million transistors (e.g. Apple A10, Intel Core i9, Qualcomm Snapdragon). A direct result of this is reflected in a significant and dramatic complexity of packages and interconnection traces contained in a single integrated circuit. Integrated circuits are increasingly dense and perform a large number of operations, requiring more current consumption and generating a rise in the circuit temperature that needs to be dissipated through the environment. The optimal placement of electronic components over a PCB can ameliorate the problem, but requires meeting multiple design objectives mostly due to the dierent power dissipation of the components, their operating temperatures, kind of materials, terminals and dimensions. It is important to notice that to approach the complexity of this type of problems, heuristics methods such as the proposed in this work, are required. Although many global companies focus in the design of electronics devices and invest in computer aided design software, specifically in Electronic Design Automation platforms known as EDA, to expedite prototype development, just a few of them have the necessary computational tools that can automatically meet all the components placement constraints. Particle Swarm Optimization (PSO) is used to address the problem through a weighted sum method for multi-objective optimization (MOO). Regardless of the deficiencies with respect to depicting the Pareto optimal set, the weighted sum method continues to be used extensively not only to provide multiple solution points by varying the weights consistently, but also to provide a single solution point that reflects preferences presumably incorporated in the selection of a single set of weights. Experiments performed were based on three main categories. Broadly speaking, the first one consisted on identifying the most important properties of PSO algorithm, its response and behavior under a dierent set of scenarios. The second kind of experiments aimed to observe and analyze which variables had more impact, and how these dominate over the rest. Lastly, the third kind cared to analyze the response of the algorithm under more complex instances of a problem. Results produced by the PSO algorithm were compared against a finite element analysis software, and finally, a general discussion was elaborated, where validity was given to the hypothesis proposed in this paper when analyzing the performance shown
- Intelligent input dimensionality reduction for thermographic computer-assisted breast cancer detection(Instituto Tecnológico y de Estudios Superiores de Monterrey) Casar Berazaluce, Alejandro; Conant Pablos, Santiago Enrique; Terashima Marín, HugoWhile traditional breast cancer thermography consists of a bidimensional image of the breast area, this study explores the possibility of reducing the size of the thermal input required by thermography to the temperature at only a few points in the breasts. That kind of information could be retrieved by locating temperature sensors around the breast, which could mean a wearable breast cancer detection implementation. In order to do this, the document first briefly summarizes and explains the methodologies used by different authors in order to implement computer-assisted thermography. Then it proposes a topological framework used to translate thermal images into a subset of points that represent what a wearable device that meets certain physical constraints would look like. This is implemented by using thermal images from 167 patients, extracting the temperature by software at 120 proposed sensor locations, 60 on each breast, and storing it in the new proposed representation. This new framework is then used to compare the methodologies used by authors in previous works, but now all under the reduced-data representation, and all using the same patient database. After finding out which kind of data representation leads to better cancer detection, the work then proceeds to use greedy search algorithms to find the subset of sensors that maximizes the predictive power of a cancer detection classifier. After the experiments, an accuracy of 88.9% was achieved using only a subset of the temperatures at only 16 locations on each breast, and an accuracy of 90.1% was achieved when combined with some statistical and spatial features computed from the temperatures. In order to evaluate the validity of these promising results, a noise and robustness test was also carried out by introducing noise into the data and studying the effects that followed. The results of this robustness test were quite relevant as well since they showed that a backward elimination search strategy can be used to remove the most noisy-prone sensors, and a forward selection strategy could potentially be used to find an optimal sensor configuration that maximizes the predictive power of the classifier. Overall, the results from this work show that thermography with reduced dimensionality achieves performances that compete with full-image thermography, and even with mammography, with the additional advantage of being implementable through a wearable device, thus not requiring the patient to visit a hospital to perform the screening. Currently the only other portable massive self-examination method is the self breast-exploration, which yields an accuracy of around 47%, which is pretty close to being a random guess. Compared to this performance, the proposed method proves to be vastly superior. It could potentially replace self breast examination as the main first-hand portable, non-invasive and non-radiative breast cancer screening tool and provide a major breakthrough in the battle against cancer by detecting the disease in the earlier stages with a greater accuracy and consistency.
- Learning temporal features of facial action units using deep learning(Instituto Tecnológico y de Estudios Superiores de Monterrey) Sánchez Pámanes, Roberto; Conant Pablos, Santiago Enrique; Campus Monterrey; Campus Monterrey; Campus MonterreyFacial expressions are an important aspect of human life and research on this topic has led to real-world technological applications. The task of recognizing facial states is involved in a collection of challenging tasks that include assisting elders and babies, as well as enhancing pedagogical exercises. Unlike categorizing faces into emotions, the Facial Action Coding System encode ambiguous expressions by analyzing small differences in the face based on muscle movements called action units. By analyzing action unit co-occurrences, human coders can virtually create any anatomically possible facial scenario that is independent of interpretation and can be used as a tool for higher-level decision processes. The automatic detection of action units in videos has recently become an interesting topic for the deep learning community since models of this area have dramatically improved the performance in image-related tasks. The state-of-the-art proposals in the benchmark database FERA17 are currently vanilla implementations of convolutional neural networks that model the occurrence of action units by ignoring their \emph{temporal features}. However, rather than being like a single snapshot, the occurrence of independent facial movements changes over time in response to information dynamically gathered from the environment, thus these deep models cannot completely capture the complex dynamic context involved in their occurrence. Researchers have engineered other deep learning methods that possess the ability to learn features across sequences of images. These procedures can be grouped into three categories, 1) methods that extend image-based architectures by using aggregation methods, or 2) recurrent units, and 3) methods that are able to process spatiotemporal features natively. They all offer the possibility of capturing AU dynamics and enhance their detection. However, their study has been frequently overlooked by the facial expression recognition community, particularly for AU occurrence detection, and up to these days, it is unclear whether deep learning models that incorporate temporal features can indeed outperform those who do not. This work analyzes the effects of incrementally adding temporal capabilities to the spatial model ResNet50 on predicting the occurrence of a single action unit of the FERA17 database. Configurations evaluated include inflating the kernels in the model to create a 3-dimensional version of ResNet50, adding a recurrent layer to encode long-term dependencies, and including the dense optical flow representation of two consecutive periods of time. Results show that adding recurrent units to a spatial model out-performs other temporal paradigms and the baseline ResNet50 by 7.4\% considering the $F_1$ score. The discoveries placed in this thesis can be utilized to better define deep learning initial implementations for projects related to facial expression recognition. Knowing the extent to which each temporal paradigm can effectively capture the dynamics inherent to AU occurrence, future research projects can be improved.