Development of a Visual Displacement and Rotation Detection System of an AGV Robot for Flat Rural Environments Using Farneback Optical Flow on Raspberry Pi 5

The use of autonomous robots in agriculture has advanced significantly, but accuracy in trajectory definition and deviation reduction remains a challenge, especially in flat rural terrains. This paper presents the development of a visual detection system for an AGV robot, whose navigation is based on optical flow analysis using the Farneback method to estimate displacements and rotations in real time from the apparent motion in image sequences. The system is given by a Raspberry Pi 5 connected by HTTP communication to the ESP32 microcontroller the movement of the robot. During the tests, it was observed an adequate detection of the camera placed on the AGV robot applying Gaussian filter and normalization to eliminate noise, with an average error in linear displacement of 0.67 cm at distances of 0.5 m, 1 m and 2 m. Meanwhile, it presents an average error in rotational displacement of 0.41° at angles of ±45° and ±90°. The system response remained stable in the face of minor changes in illumination and soil texture, which evidences the viability of the proposed visual control scheme. In the same way, this developed method presents better results than multisensor systems made for AGV robots in rural environments.