https://repositorio.ufu.br/handle/123456789/19216 2026-09-24T20:58:45Z https://repositorio.ufu.br/handle/123456789/50360 Title: Aplicação de estabilidade em uma clínica odontológica de pequeno porte Abstract: Small dental clinics usually focus on clinical excellence while neglecting support processes, which leads to material waste, delays and team overload. This work proposes the implementation of the Lean Healthcare operational stability concept in the support sector of a small dental clinic, focusing on the storeroom and the sterilization center. The research, applied in nature and with a quali-quantitative approach, was conducted through action research, with participant observation, inventory audit, time study and photographic records, structured by the A3 Report and three Ishikawa diagrams adapted to the 6Ms, one for each measured effect. The diagnosis identified losses of BRL 7,190.00 per year in expired materials, a 40-minute room setup time and up to five interruptions per surgery to search for missing items, in an operation of approximately 130 full-setup surgeries per year. The implemented countermeasures (5S with visual management, FIFO system, standardized surgical kits and a standard operating procedure for room assembly) reduced setup time to 15 minutes (62.5%), eliminated surgical interruptions, stopped the disposal of expired materials and removed an average monthly expense of BRL 46.00 on emergency freight, freeing about 50 hours per year of the bottleneck resource. The stabilization of the 6Ms proves to be a feasible, low-cost prerequisite for continuous improvement in dental services. 2026-08-14T00:00:00Z https://repositorio.ufu.br/handle/123456789/50358 Title: Implementação de robô móvel para rastreio de pessoa com visão computacional Abstract: The development of an autonomous robot capable of tracking people integrates knowledge areas such as computer vision, control systems, and communication protocols. In this con- text, this work aims to develop and implement a mobile robot that performs human tracking in unstructured environments. The robot architecture is based on a high-level processing unit, implemented on a Raspberry Pi, which executes the computer vision algorithm and the control logic; the low-level processing unit, based on the ESP32, interfaces with the power system and controls motor actuation through an H-bridge; finally, the system actuators consist of two direct current motors driven in a differential and independent manner. The computer vision algorithm, developed in Python, processes the image stream from acquisi- tion to the extraction of the required information. Upon acquisition, the images are converted into numerical matrices using the OpenCV and NumPy libraries to feed the YOLO11 (Ultra- lytics) neural network. This network is responsible for detecting people and extracting the parameters of the image bounding box, using the horizontal center coordinate and the height of the bounding box to perform, respectively, orientation correction, keeping the robot aligned with the target, and distance control, stabilizing the robot within a resting range. These values are normalized and undergo stochastic filtering using the Kalman filter, which estimates the variables and reduces detection noise. The estimated variables are then provided to the finite state machine, where the robot’s operational states are defined. The velocities are defined and transmitted via the MQTT communication protocol over a local network; this information coordinates the differential traction system controlled by an H-bridge and direct current motors, applying hysteresis, deadband, and intermittent actuation techniques to mitigate static friction and chattering. The experimental results demonstrate the robot’s ability to maintain target tracking with stable behavior, operating within the required time to follow a person walking through the environment. This result highlights the feasibility of integrating computer vision techniques with embedded control in autonomous mobile robotic systems. 2026-03-17T00:00:00Z https://repositorio.ufu.br/handle/123456789/50054 Title: Melhorias estruturais de um módulo robótico acoplado em um drone para instalação de esferas de sinalização por alavanca Abstract: This work presents structural improvements to a mechatronic module designed for the remote installation and removal of lever-type aerial marker balls on high-voltage power lines, aiming to reduce workers' exposure to the risks associated with this operation. The development consisted of the mechanical redesign of a previously developed structure, focusing on weight reduction, improved static balance, and the restriction of undesired degrees of freedom during system operation. The implemented modifications included structural changes and adaptations to the mechanism responsible for opening the aerial marker ball. After the improvements were implemented, functional tests were conducted to evaluate the module's performance under laboratory-simulated conditions. The results demonstrated improvements in the mechanical behavior of the structure and identified limitations related to the system's reliability, highlighting the need for further optimization before its practical application. 2026-07-28T00:00:00Z https://repositorio.ufu.br/handle/123456789/50049 Title: Emprego de visão computacional para identificação de comportamento suspeito no varejo Abstract: This work addresses the chronic problem of inventory losses due to theft in the retail sector, aiming to empirically compare the performance of two known deep learning architectures for the automated detection of suspicious behavior in videos: a Three-Dimensional Con- volutional Neural Network (I3D) and a Vision Transformer (TimeSFormer). To conduct the study, the methodology involved building a proprietary dataset through the aggrega- tion, filtering, and segmentation of samples from public datasets, subjecting the models to sixteen experimental configurations. These experiments varied the weight fine-tuning strategy, temporal resolution, and the use of optical flow images combined with standard RGB inputs. The evaluation also included a simulation of continuous monitoring on real, unedited videos using a sliding window inference approach. The results demonstrate that full fine-tuning of all network parameters is a strictly essential step for convergence and learning in both architectures. The Vision Transformer achieved the best generalization capability and false alarm control on the test set, attaining perfect precision; however, it requires substantially higher computational investment and exhibits high inference time. In contrast, the Convolutional Neural Network showed a more conservative predictive behavior with a greater tendency toward false positives, while standing out for its high efficiency and low latency. It is concluded that practical deployment requires a direct tech- nical trade-off between the high selectivity of attention-based models and the scalability and speed of convolutional networks. Additionally, both systems share critical vulnera- bilities and detection failures when faced with severe occlusions and subtle movements in retail environments. 2026-05-21T00:00:00Z