This work presents a comparison between techniques for recognizing inorganic molecular structures from hand-drawn sketches. The approach used combines Convolutional Neural Networks (CNN) and the Histogram of Oriented Gradients (HOG) method to classify and recognize these sketches. For the tests, we used images drawn by high school students. The comparison of the results obtained with the CNN and HOG techniques was carried out in detail. Additionally, we developed an application capable of generating three-dimensional visualizations of the molecular structures, allowing their representation in a virtual environment. This 3D visualization provides a more intuitive understanding of the molecular structures. This work highlights the effectiveness of machine learning technologies in education, by offering an accessible and interactive educational tool that complements traditional chemistry teaching. In this way, it provides students with a deeper understanding of the physical and chemical properties of inorganic molecules, enriching the learning process and making it more engaging.

Neurodegenerative diseases (NDDs) cause, among other symptoms, motor impairment. Given the incurable nature of the NDDs, several studies have investigated gait using artificial intelligent models as non-invasive alternative methods to assist in the diagnosis of these diseases. This work proposes a novel method using an Encoder-Only Transformer to detect NDDs, a multi-classification task, through gait signal analysis. The approach comprises data preprocessing, windowing technique, a modified transformer architecture and cross-validation evaluation. The results indicate the transformer-based architecture can be a promising alternative to accomplish this goal.

This paper presents the "Letter Explorers" platform, an educational tool designed to support preschool children in their writing acquisition process. The platform was designed based on recognized pedagogical guidelines, structuring its activities into progressive levels of difficulty. The main focus is on learning letters and encouraging independent writing, respecting the motor and cognitive development of young children. Its features include real-time visual and auditory feedback, which helps correct and reinforce learning, as well as the possibility of personalizing activities based on the child’s name, a factor that contributes to motivation and engagement. The proposal aims to serve, especially, educational contexts in the interior of emerging countries, where access to quality digital solutions may be limited. Preliminary results indicate the potential of the tool to promote an interactive, inclusive and contextualized environment for learning to write in preschool education.

Tuberculosis (TB) remains a major global health crisis, disproportionately affecting vulnerable populations. Despite advances in artificial intelligence (AI) for chest X-ray (CXR) analysis, these tools have limited impact in low-resource regions due to inadequate infrastructure, specialist shortages, and high equipment costs. This study proposes a practical AI-based approach using optimized binary phase pattern congruence (BPPC) feature selection to distinguish between TB cases and healthy individuals. Its lower computational requirements and costs make it particularly suitable for vulnerable regions. We experimented with multiple CXR databases and segmentation scenarios using optimized feature selection. Results outperform existing literature, achieving a minimum area under the curve (AUC) of 97.64%, showing potential to enhance CXR analysis and assist specialists in TB diagnosis.

Assistive technologies enable individuals with special needs to access resources that would otherwise be inaccessible due to their condition. In this context, people who are blind or visually impaired face daily challenges related to mobility, communication, and, most notably, access to education in specialized fields. In this work, we perform a systematic literature review that aims to identify and map assistive technologies designed to support programming education for individuals with visual impairments. The review was conducted across three academic databases and resulted in the selection of 22 articles that address the research questions posed in this study.

Depth estimation plays a crucial role in understanding spatial relationships within natural scenes, enabling applications in 3D modeling, robotics, and environmental monitoring. This paper presents a comparative study between the Depth Anything model—a monocular depth estimation framework—and LiDAR sensors in forest environments. A real-world dataset of 4,613 frames captured from a eucalyptus farm in Açailândia, Maranhão, Brazil, was used for evaluation. Depth Anything was trained using a large-scale dataset with 1.5 million labeled and 62 million unlabeled images. The error metrics used in this study include MAE (Mean Absolute Error) and RMSE (Root Mean Squared Error), achieving total values of 0.1096 meters and 0.1328 meters, respectively. These results demonstrate high alignment with LiDAR measurements and robustness in complex environments. Furthermore, the analysis includes a per-tree evaluation and statistical distribution through boxplots, confirming stable and consistent predictions. This study suggests the potential of combining monocular models with traditional sensors to enhance depth estimation for forest management and biodiversity monitoring.

This paper presents a computational approach for the machine translation of comic book texts, aiming to promote linguistic accessibility, particularly in emerging countries. The proposed system performs end-to-end processing: it detects text regions, extracts content using Optical Character Recognition (OCR), translates the text into the target language, and reinserts the translated content into the original image, preserving its visual structure. Built with open-source libraries, the system is lightweight and suitable for low-resource computational environments. The methodology was validated through an experiment involving 1,000 pages of comics in four languages—English, French, Spanish, and Japanese. Results demonstrated high accuracy in the OCR and translation stages for Latin-based languages, and satisfactory performance for Japanese, despite its right-to-left reading layout and ideographic characters. The tool also showed potential as an assistive reading solution, with applications in educational and inclusive contexts. This work contributes to the development of accessible technologies aligned with the United Nations Sustainable Development Goals (SDGs), particularly in promoting quality education and reducing inequalities.

This work presents a method for interacting with video players through real-time hand gesture recognition, implemented using the MediaPipe library and the Random Forest algorithm. The main goal was to develop a lightweight and accessible system capable of controlling playback, time navigation, and volume adjustment without requiring physical contact with the screen. The prototype was developed for Android devices and evaluated through an experiment conducted under real usage conditions, where users interacted with the system on their own mobile devices. Usability was assessed using the System Usability Scale (SUS), which resulted in an average score of 83.38, classified as "Good." The user experience was further evaluated using the short version of the User Experience Questionnaire (UEQ-S), which yielded "Good" results in the pragmatic dimension and "Excellent" in the hedonic dimension. The results indicate that the system is intuitive, functional, and pleasant to use, even among users with little prior experience with gesture-based interaction. The model demonstrated good generalization capabilities, even with a reduced training dataset. The proposed approach proved to be viable and promising for future applications in various contexts, such as entertainment, education, and accessibility.

The Brazilian capital market presents unique challenges for portfolio optimization due to its volatility and information asymmetries. This paper proposes a Deep Reinforcement Learning (DRL) framework that integrates fundamental indicators, Portuguese-language news sentiment (via Gemini Pro), and market data (prices, volume). Five DRL algorithms (A2C, PPO, DDPG, TD3, SAC) were trained and evaluated across three feature scenarios using performance metrics such as Sharpe Ratio, Annual Return, and Maximum Drawdown. News sentiment classification and entity extraction were performed using Gemini Pro LLM. Statistical validation over 44 executions, including Shapiro-Wilk and Kruskal-Wallis tests, showed no significant differences among DRL methods. However, all DRL approaches outperformed the Ibovespa index and uniform Buy-and-Hold benchmark, highlighting the value of combining DRL with localized and diverse data sources for portfolio optimization in emerging markets.

This paper presents an accessible method for solfège instruction tailored to blind and visually impaired students, using the Ableton Live software. The proposed approach allows users to create custom MIDI instruments by recording the spoken names of musical notes, which are then mapped onto MIDI files to replace traditional note sounds. This personalized method promotes greater familiarity with musical notation and supports the development of music reading skills without relying on visual cues. The process involves converting sheet music into a MIDI file, identifying the main octave, recording the note pronunciations, and assigning them to a sampler in Ableton Live. Once configured, the customized instrument enables real-time auditory feedback as the notes are played, facilitating a more inclusive and autonomous learning experience. This approach addresses common barriers in traditional solfège instruction for blind learners, such as limited access to accessible materials and the visual nature of many teaching methods, by leveraging audio-based interaction and widely available digital tools.

Many investors still rely on their emotions as their primary guide for asset allocation, blindly following one or two news sources that may or may not be an actual synthesis of the market, without any mathematical formulation to support them, even if slightly. This work aims to conduct a Systematic Review using Kitchenham’s protocol to understand the state of the art regarding the combination of Portfolio Optimization techniques and Artificial Intelligence in the context of trading assets, with a special focus on Sentiment Analysis techniques. This was achieved through the definition of search keywords used in 4 different major research databases and selection through inclusion and exclusion criteria. A total of 384 articles published between 2019–2025 were identified, among which 27 articles were selected that fit all the selection criteria. The research questions address the specific techniques employed for optimization, with the most proliferated being Deep Reinforcement Learning.

This work investigated the usability of two interaction techniques in low-cost virtual environments using smart-watches: on-screen continuous gesture recognition and the MDDTW algorithm for touchless gestures. While continuous gesture recognition requires direct interaction on the device's screen, MDDTW allows users to perform gestures in the air without the need for physical touch. Although users initially preferred continuous gesture recognition, the results revealed that MDDTW achieved slightly higher scores. This underscores the crucial importance of considering user experience in the development of new technologies. The comparative analysis between the two approaches contributes to the design and implementation of interactions in accessible virtual environments, especially concerning the integration of physical gestures as part of the interaction. This work contributes to understanding the factors influencing usability in virtual environments and highlights the need for user-centered approaches in designing interactive technologies.

In this work, we propose a method for recognizing percussive gestures using smartwatches with accelerometers and the MDDTW algorithm, incorporating an activation threshold to identify the beginning of gestures. We developed a system that allows simulating drum percussion in a musical duet, providing an interactive and engaging experience for users. Our method utilizes an activation threshold based on the value of gravity to identify the onset of percussive gestures, enabling precise and efficient detection of user movements. We conducted a controlled experiment where participants were instructed to perform predefined percussive gestures, which were captured by the smartwatch sensor and processed by the system. The results demonstrated good accuracy, with a consistent recall rate, indicating the system's ability to correctly identify performed gestures. The Fl-score, as a combined measure of precision and recall, confirmed the overall good performance of the method.

Traditional paper-based children’s spelling assessments were hampered due to Covid-19 because existing technologies did not provide strategic signals to teachers, such as the child’s handwriting direction and how they read what they write. Our project emerged as a novel method to assess children’s spelling by touchscreens in this context. Hence, this paper aims to extend community knowledge concerning children’s experience and perception of handwriting spelling on tablet devices. The experiment consisted in presenting three handwriting methods (paper and pencil, finger and pen writing) and was conducted with eight Brazilian children between 4.5 and 7 years old. In addition to observation, in our experimental protocol we adopted the Fun Sorter, Again-Again Table, and the Smileyometer as evaluation tools. Our results show children were excited about handwriting using a touch pen on the tablet. Most of them even revealed they prefer the pen tablet mode to the traditional paper and pencil mode. However, the majority of children did not feel comfortable writing by finger, and it required more time than other methods. Furthermore, we observed child’s handwriting using finger looks different when compared to paper and pencil, while the tracing using a touch pen is similar to the registration produced on paper.

Tuberculosis is a severe and contagious lung dis-ease that kills about 1.5 million people worldwide. One of the ways to combat this disease is by tracking, detecting, and iso-lating the infected. In this sense, chest radiography (CXR) is an effective alternative for this task, given its high availability, low charge, and quick response. Thus, considering the importance of this topic, our work proposal is a machine learning method for tuberculosis detection in CRXs. Our method combines local binary patterns (LBP) feature extraction and a feature selection wrapper algorithm by Monarch Butterfly Optimization (MBO) with an evaluation KNN classifier. The results are compared to a reference work on various metrics and show 90.33 % and 92.41 % accuracy and the area under the ROC curve, respectively. Our proposal is a solution that combines performance, reduced computational cost, and simplicity of implementation, composing a viable and aligned alternative to the Internet of Things (IoT) solutions.

In this work, we explore machine learning to ob-tain sentiment indicators from financial market text messages collected from Twitter and news in Portuguese. A statistical analysis was carried out with Sperman’s correlation coefficient between sentiment indicators and actual variables in the Brazilian financial market. For sentiment analysis, the MaxEnt and CNN models provided F1-scores of 85% and 96%, for tweets and news, respectively. A result shown a moderate correlation (Cohen’s scale) between some variables such as, amount of tweets and sentiment of the news; market variation and tweet sentiment; amount of retweets published and sentiment of the news; trading volume and tweet sentiment. Moreover, a very large correlation was identified between the amount of negative tweets and retweets, leading to the belief that pessimism is often propagated.

Since vaccination started, the COVID-19 scenario has improved. On the other hand, although the number of deaths has significantly dropped, the number of new cases is still a concern. Thus, patient tracking and follow-up are essential tasks, and chest X-ray examination is the first-order tool. While several studies using CXR and computing have been developed, they did not translate into clinical applications yet. One of the reasons is the computational effort required to run huge deep learning models and its high cost to be adopted in community clinics. Therefore, this work proposes a lightweight (few computational resources needed), fast (training and inference time), and reasoned solution for automatic COVID-19 detection and assessment of its severity. Our method is based on extracting features by Binary Pattern of Phase Congruency (BPPC) in segmented CXR images. Radiomic features are extracted from the segmented CXR image, and an SVM-based selection process is used to build two models of a shallow Feed-Forward network. The results surpass previous studies, with an average accuracy for COVID-19 detection of 98.71%. For images without evidence of infection but with a positive PCR test, an accuracy of 94.74% is reached. In a second task, the severity level of COVID 19 is estimated with an AUC of 98.92%. This high performance helps improve the speed and accuracy of diagnosis and severity assessment of COVID19 infection, proving to be a viable option in transitioning from a research field to a clinical environment.

Governments, civil society, health professionals, and scientists have been facing a relentless fight against the pandemic of the COVID-19 disease; however, there are already about 150 million people infected worldwide and more than 3 million lives claimed, and numbers keep rising. One of the ways to combat this disease is the effective screening of infected patients. However, COVID-19 provides a similar pattern with diseases, such as pneumonia, and can misguide even very well-trained physicians. In this sense, a chest X-ray (CXR) is an effective alternative due to its low cost, accessibility, and quick response. Thus, inspired by research on the use of CXR for the diagnosis of COVID-19 pneumonia, we investigate classical machine learning methods to assist in this task. The main goal of this work is to present a robust, lightweight, and fast technique for the automatic detection of COVID-19 from CXR images. We extracted radiomic features from CXR images and trained classical machine learning models for two different classification schemes: i) COVID-19 pneumonia vs. Normal ii) COVID-19 vs. Normal vs. Viral pneumonia. Several evaluation metrics were used and comparison with many studies is presented. Our experimental results are equivalent to the state-of-the-art for both classification schemes. The solution’s high performance makes it a viable option as a computer-aided diagnostic tool, which can represent a significant gain in the speed and accuracy of the COVID-19 diagnosis.

In methodologies that make use of the remote sensing images obtained by orbital sensors, it is very common the application of resampling methods with the adaptation of images contained bands with different spatial resolutions, for example, the Sentinel-2 sensor, with thirteen bands, four with a resolution of 10m, six of 20m and three with 60m. In this way, to calculate some vegetation indices, the difference of spatial resolution among bands does not allow index calculation, requiring the application of resampling. In the literature, there are several techniques, but the effects derived from that pixel transformation have not been explored much when related to sugarcane classification. Thus, this paper applies different resampling methodologies focused on remote sensing, verifying the effects of each transformation in the vegetation indices calculation to perform sugarcane varieties discrimination. It was possible to observe little variation in accuracy amid the applied methods, showing little influence in the process to identify sugarcane varieties. Thus, the use of indices related to chlorophyll demonstrated great potential for the purpose of discriminating/classifying sugarcane, presenting alternative vegetation indices to be applied for this type of purpose.

Management of agricultural production and rural activities has been supported by recognizing machine learning patterns and algorithms, as in the automation of leaf analysis. However, leaf border damage compromises leaf structures, making it difficult to estimate the lost contours. Effects caused by predatory insects are difficult to be monitored by inspection processes, and the harmful results caused by them can deteriorate the performance of machine learning models. In this sense, plant leaves that are not fresh or intact are avoided. Consequently, the number of samples for use in training steps is reduced, leading to problems of data balancing and limited generalization models. This study presents an automatic method for reconstructing an injured leaf at a probable stage before defoliation. Thus, the reconstruction of damaged leaves can be used to maximize the number of samples in the plant species classification processes and provide visible results for the agronomic analysis of regions of occurrence of leaf damage and the components of the primary leaf structure affected by predatory insects. Based on the experimental results, we conclude that the proposed approach can accurately delimit the injured leaf silhouette and restore the leaf regions affected by herbivory attacks.

This work aims to investigate, by means of a Systematic Literature Review, to evaluate the current state of the use of artificial intelligence in automated pupillometric technology and its application in helping to diagnose diseases, to identify the methods and equipment used and propose case new equipment based on computer vision is feasible. We also investigated the accuracy of methodologies and equipment that use computerized pupilometry to identify pathologies or disorders, as well as the viability and usability of existing pupilometers. In this sense, creating a pupilometer capable of stimulating and varying wavelengths, providing an interface to preview the exam, and embedding the classification algorithms is a great challenge. In this systematic review of the literature, we consider publications from the last ten years (2010 - 2020) indexed by seven solid scientific databases. The review identified a vast amount of work on pupillometry; however, a small amount related to the construction and viability of a pupilometer with an embedded system, easy to use and with a preview interface. Having identified this, we propose a new methodology for the construction of the pupilometer as well as the algorithm for extracting the characteristics through pupilometry.

Chest radiography (CXR) is one of the first choices in epidemiological analyses such as tuberculosis, cancer, pneumonia, and, recently, COVID-19. It provides crucial information for decision making, treatment, and monitoring the evolution of clinical cases from small to high complexity. Thus, it is a valuable source of information for the study, training, research, and development of computational support to medical diagnoses. In this work, we introduce a new method for chest X-ray adjustment to identifying and correcting radiographic images orientation. So, they can be automatically rotated to a standard position. Our proposal uses structural characteristics and statistics of pixel intensity patterns of CXR images. Divided into three steps, our method begins with the preparation of the photos, followed by a feature extraction strategy, and it ends with the X-ray image orientation identification. We use three different databases that include pediatric and adult radiographic imaging. A result showed 99.4% accuracy in the databases in our experiments. The code prepared by the authors is publicly available.

Remote sensing techniques by satellite imagery have been widely applied in various fields of agrarian sciences due to allowing real-time information, allowing data retention in a given region without the need for displacement, avoiding costs, and also enabling the creation of more efficient methods for the task of monitoring crops. In special to remote sensing applied to sugarcane varietal identification, the possibility of discrimination among the varieties is important due to allows the monitoring of the crop growth concerning characteristics by plants, measures controls, and the preservation of copyright of developed varieties. Among the researches involving studies with sugar cane regarding varietal identification, the purpose of the paper implies to present a review of the literature, conferring methods, and checking state of the art about the subject of discrimination of sugarcane varieties by remote sensing.

Assessment of children's spelling development stages is an activity frequent in literacy classrooms. Usually, teachers adopt dictation sessions, using a paper-and-pencil based method, which has to be conducted individually with each child. As there are many students in a class, the activity becomes laborious to be offered frequently, pushing teachers to opt to a sub-optimal amount of tests. Moreover, the number of tests to be performed in person is now limited. In this context, we aim to develop a method for conducting automated word dictation sessions for an in-person and remote assessment. Besides, our proposal aims to support teachers, parents, and other literacy professionals in identifying children's spelling development stages. In this paper, we report the conception of our computational artifact through the Design Science Research Methodology. After reviewing the existing studies, we developed a high fidelity prototype and evaluated the concept during a focus group discussion with literacy teachers. The collected qualitative result indicates the feasibility and utility of our approach, and evident limitations in existing apps to assess children's spelling, contributing to future research in this area.

Smartwatches are wearable devices of emerging mobile technology and are increasingly in people's daily lives. They allow you to perform various tasks; however, the interaction with these devices still proves to be a significant challenge. Several studies carried out interaction studies with smartwatches using gesture recognition. Thus, this work presents a comprehensive Systematic Review of Literature (SRL) on interaction with smartwatches using gesture recognition, showing what has already been developed and what is state of the art. We searched in four databases with relevant scientific scope: ACM Digital Library, IEEE Xplore Digital Library, Science Direct, and Scopus. Work on this theme is diverse; gesture recognition is used to perform operations on smartwatches itself or even control other devices or virtual environments. We hope that this work will provide a rich base on the methods developed and that it will become a source of research for future researchers and that can support the development of other methods and applications.

Identifying the phase or stage of children's spelling development is a regular activity in literacy classes. Usually, teachers assume some developmental theory and perform paper-and-pencil based tests. Existing digital tools often do not consider any of these known theories, nor do they capture the child's handwriting. Therefore, some teachers prefer to continue applying the tests manually. Accordingly, this study's research question is how to promote the performance of spelling tests on mobile devices, simulating the interaction between manual tests and the theoretical models already used by teachers. Through the Design Science Research Methodology (DSRM), we propose a method to apply child spelling tests in an automated way. In this work, we present the results of the child's interface usability evaluation in the developed computer artifact, using guidelines of Touchscreen Interaction Design Recommendations for Children (TIDRC). The results indicate adequacy to the recommendations of 88% of items in visual and audio features (cognitive dimension), 75% in the physical dimension, and 47% in socio-emotional dimension. These results are promising and relevant compared to previous studies that evaluated apps using the TIDRC framework.

This work proposes the development of a method that allows the use of smartwatch as a control for interactive films using continuous recognition of gestures and conducts a case study with the film "Black Mirror: Bandersnatch". We developed two prototypes to interact with interactive movies. The first one uses gesture recognition on the smartwatch screen. Thus, we created a set of gestures composed of straight lines that represent the actions in the interactive movie. In this prototype, the recognition of gestures is performed by the algorithm of continuous recognition of gestures, in this way, a gesture can be recognized before the user finalize, and action is sent quickly to the movie. The second prototype uses a touch of pressure in the smartwatch to control the film. Thus, the actions to be performed in the smartwatch that represents the actions in the interactive film is proposed, the recognition of the gestures is determined by the change of the values obtained by the accelerometer. The prototypes communicate with a simulation service responsible for running the movie. The prototypes were used in a study with users, as well as in usability and experience tests, and the results show that the method has the potential to be used in everyday users to control interactive films efficiently and effectively.

Analyzing human pupillary behavior is a noninvasive and alternative method for assessing neurological activity. Changes in this behavior are correlated with various health conditions, such as Parkinson's, Alzheimer's, autism and diabetes. Examining pupil behavior is a simple, low-cost method that can be used as a complementary diagnosis in comparison with other neurological evaluation methods. This approach is made by recording the pupillary behavior against light stimuli and measuring the pupil diameter through the video. The relation of pupillometry with digital image processing creates a dependency for computer vision based systems. Therefore, this paper presents a systematic review of the literature (SRL) conducted in order to analyze the progress of pupillometry systems based on computer vision. The main goal was to establish the state of art and identify possible gaps.

A disparity map is a key component of stereo vision systems. Autonomous navigation, 3D reconstruction and mobility are examples of areas that use disparity maps as an important element. Although much work has been done in the stereo vision field, it is not easy to build stereo systems with concepts such as reuse and extensible scope. In the present paper, we contribute to reducing this gap by presenting a software architecture that can accommodate different stereo methods through a new standard structure. Firstly, we introduce scenarios that illustrate use cases of disparity maps, and we show a novel architecture that foments code reuse. A Disparity Computation Framework (DCF) is presented and how its components are structured regarding compartmentalization are discussed. Then, we introduce a prototype that closely follows our proposal, and we describe some test cases that were performed. We conclude that the DCF can satisfy different on-demand scenarios and that it can support new stereo methods, functions, and evaluations for different applications without much effort.

Medical X-ray images are an important and valuable source of studies and diagnoses for diseases with low cost besides its high availability. However, radiological images are subject to degradations related to low contrast and presence of noise. Based on this finding, this article presents a simple but efficient enhancement method for these images with the objective of contrast gain and noise removal. The proposed method (MP) consists of a sequence of interactive steps. Start from the step of double precision conversion and end with removing impulsive noises. An evaluation with the PSNR, Entropy, AMBE, and IQR indicators was performed, besides gain check on the thresholding process and the histogram characterization. The evaluation was conducted on three different datasets in a total of 1409 images chest X-rays. The results compared to others known in the literature proved to be promising and put it as an interesting alternative in the process of enhancement medical X-ray images.

Since 1946 methods for sensory substitution of vision has been studied; however, half a century after the beginning of this line of research, this keep been a massive problem in a world with about 50.6 million people with irreversible blindness. This research presents how self-help devices for visually impaired are approach in recent years and proposes a new approach based on object recognition with deep learning. Through it, it is possible to perceive the trends in this line of research, how devices obtain information from the environment, how they interact with users, and other aspects - pointing essential factors to all those who research or wish to study this area.

In this study, we explore some possibilities of using aerial images captured by Unmanned Aerial Vehicles (UAV) and discuss the benefits of using them in the context of intelligent agriculture. A novel method that supports the detection and segmentation of tree crowns, the delineation of shadows, and which shows the direction of sunlight is presented. It uses simple observation strategies and commonly used digital image processing techniques such as visual color enhancement and perception, morphological operations, and segmentation based on a region growing method. The proposal is evaluated using a dataset with different types of crop areas and pasture lands. The results indicate that the proposal can effectively deal with the detection and segmentation of elements of interest in the scene, as well as the indication of the right side of the light source.

Computer vision is an area have proven to play an essential role in urban and rural applications like medical, agriculture, and remote sensing. The use of image processing methods for simulating the visual capability of robots plays a crucial role in the consolidation of smart farming. The understanding of the complexity of outdoor environments, where the robot performs its task, is an essential issue for the development of efficient processes of autonomous mobility, especially in areas with uneven illumination, unpredictable weather conditions, and different color shades. In this study, we present a new method to detect and segment tree trunks from unstructured environments where natural properties such as lighting and terrain shape form a variety of non-controlled conditions. We prepared a dataset with stereo image pairs and ground truth maps to calculate disparities and to evaluate the proposed method in the application of smart farming. The results show that the presented approach can segment trees with high precision, which is an important step in calculating the disparity of external components by systems that use the stereoscopic view.

Microblogs have increasingly been used by the crowd to post their thoughts and speeches about everything. Thus, one of the themes is the stock market that is exploited by many researchers. Although obtaining indicators of stock market dynamics through online social networking has been gaining the attention of academia and the business world, there are many questions to be analyzed about their effectiveness. This work presents the development of a simulator for buying and selling stocks based on microblog data. Therefore, we collected tweets about the Brazilian stock exchange market, produced indicators using sentiment analysis and performed a set of heuristics for decision making. The first technique is the composition of the decision-making strategy for buying and selling stocks composed of pure logic, tweets volume thresholds, profit objective and technical analysis in the stock exchange. The contribution of this work is a decision-making architecture using Twitter's data as an index of future expectation about the social mood that may change the market behavior. As a result, it has pointed to attractive profits for Brazilian market actions and many issues that can be analyzed and improved. Our study showed that it is possible to obtain market dynamics information on the Twitter social network and this information could be used to compose stock buying and selling strategies.

As an emerging technology that combines mobile and wearable markets, smartwatches are finding their place on consumers' daily lives. They allow tasks that used to be performed only by smartphones and tracking devices. Despite the increasing interest on them, a task that is still not fully covered by these devices is text entry, mainly due to their reduced screen size. Researchers have been working hard on solutions for this issue in the past years, and a number of methods for interactive text entry with smartphones now exist. The aim of this paper is to present a systematic review on these methods, showing what has been developed and what is the performance of the current state of art of the technology. The review focused on four databases. After applying a large selection criterion, it resulted in twenty-six approaches, which helped to answer questions that grounded this work. We hope to deliver a rich and useful foundation about methods, results, challenges and opportunities and to support new research on smartwatches.

This work proposes the development of a method for smartwatches that allows to control platform games using continuous recognition of gestures and conducts a case study as the game Super Mario World. Uses a set of gestures based on geometric shapes to send actions to the game. Gesture recognition is performed by the algorithm of continuous gesture recognition, as it is able to recognize a gesture before being finalized, allows an action to be performed quickly, improving feedback. The recognition process was paralleled to improve performance. A technique has been developed that allows the execution of several gestures in sequence, without the need for a signaling that a gesture has been finalized or initiated. It was also created a technique that allows the sending of special commands to the game using the pressure applied on the screen by the player. A prototype for smartwatches was developed that communicates with an emulation platform installed on a Raspberry PI 3. A user experiment was performed as well as usability and experience tests. The results show that the method has the potential to be used effectively and effectively by players.

As a digital image provides such information about a scene, a disparity map can be yielded by means of stereo images. This topic was exhaustively surveyed but it remains one of the most important branches in both computer vision and machine vision. Most algorithms are organized in a pipeline that starts with a matching cost step and ends with a disparity refinement. This paper provides a simple but an effective method to adjust a disparity map in a more appropriate configuration, i.e. it presents a disparity refinement technique. It is based on an assumption that most disparities in a region point to a correct disparity value for this area. To develop the methodology, we use image segmentation and support weighted windows. By performing an evaluation, it shows that this method can increase the robustness of a raw disparity map even with a lot of noisy parts.

Emerging technology on mobile and wearable market, smartwatches have embedded movement sensors whose potential is yet to be fully explored. This paper proposes an interaction method with smart TVs via gestures performed by person's wrist using a smartwatch. Detailed architecture and implementation for a complete prototype, named Wrist Player, is presented. A user study is also conducted, in order to evaluate the prototype performance and the user's interest on the proposal. Results show that the method works very well, with participants reporting having a good experience with the prototype. We present our insights on the concept, challenges faced in our research and ideas for future studies.

This work proposes a method that allows the entry of text in smartwatches using gestures based on geometric forms. For this it is proposed the development of a prototype capable of inserting a letter with no more than two user interactions. Gesture recognition is performed using the incremental recognition algorithm. A set of gestures with lines and curves were created to be recognized by the incremental recognition algorithm, generated from the reduced equation of the line and the reduced equation of the circumference, respectively. After recognizing the gestures, they are sent to a classifier Naïve Bayes which is responsible for predicting the letter that will be inserted. The Naïve Bayes classifier was trained with a user gesture base that drew all the letters of the alphabet using only the gestures available in the set presented to them. Using the gesture base and the classifier Naïve Bayes a prototype was developed for smartwatches that automatically suggests the most likely letters to be inserted. The prototype was used to perform an experiment, during the experiment the users inserted the five most frequent letters and the five less frequent letters of the English language. The results of the experiment show that the prototype is able to recognize a letter with at most two interactions between the user and the smartwatch. The analysis of the usability and experience test shows that the prototype has generalized potential for use, since it allows the entry of text with up to two interactions and with a 100% hit rate for the most frequent letters and 95,14% For less frequent letters.

This work extends PneumoCAD, a Computer-Aided Diagnosis system for detecting pneumonia in infants using radiographic images, with the aim of improving the system's accuracy and robustness. We implement and compare five con-temporary machine learning classifiers, namely: Naïve Bayes, K-Nearest Neighbor (KNN), Support Vector Machines (SVM), Multi-Layer Perceptron (MLP) and Decision Tree, combined with three dimensionality reduction algorithms: Sequential Forward Selection (SFS), Principal Component Analysis (PCA) and Kernel Principal Component Analysis (KPCA). Current results demonstrate that Naïve Bayes classifier combined with KPCA produces the best overall results.