Melhoria de navegação para um robô de trekking por meio de processamento de imagem

Abstract

Autonomous navigation of mobile robots, especially in indoor environments where technologies such as GPS face significant limitations, represents one of the major challenges of contemporary robotics. In the context of the indoor Trekking Robot competition, the development of an embedded computer vision system was proposed. The objective was to develop, implement, and validate the feasibility and performance of an embedded computer vision system for the Patobots team’s Trekking robot, using the detection of signaling cone colors to generate real-time trajectory correction commands. To this end, a Raspberry Pi 4 was employed as the processing unit, coupled with a USB camera and an MG90S servo motor, integrated with algorithms implemented in Python using the OpenCV library. The vision system was designed to identify colored signaling cones present along the competition track by using color-space conversion to Hue, Saturation, and Brightness (HSB), range-based threshold segmentation, and contour extraction. Based on this information, the prototype was able to compute the target’s center and apply active tracking strategies in real time. During system testing, the approach achieved average rates of up to 70 frames per second, demonstrating sufficiently low latency for dynamic control applications. The control actions derived from vision are transmitted via serial communication (Universal Asynchronous Receiver/Transmitter – UART) to the STM32 microcontroller responsible for locomotion, establishing the architecture for a future hybrid control strategy that combines inertial and visual data. These results demonstrated the system’s capability to accurately identify targets and generate the steering commands required to align the robot with its objective. Full validation of the navigation system in the field was not carried out due to scheduling constraints, with integration and sensor fusion identified as the main directions for future work. The project presents a high-performance and well-documented vision system that provides a solid and promising foundation for the full future implementation on the Trekking Robot.