Дослідження ефективності застосування візуальних маркерів у задачах локалізації всередині приміщень

UKR: В роботі проведено дослідження ефективності застосування візуальних міток, таких як маркери ArUco, ARToolKitPlus та QR-коди, для локалізації автономних роботів, що переміщують вантажі на складських приміщеннях. Розроблено імітаційну модель у середовищі Matlab, за результатами моделювання отримано оцінку втрат в умовах помилок розпізнавання міток. Результати роботи можуть застосовуватися при створенні та аналізі ефективності алгоритмів локалізації на основі візуальних міток всередині приміщень.

ENG: This paper examines the effectiveness of using visual markers such as ArUco, ARToolKitPlus and QR codes to localize autonomous robots moving goods in warehouses. Using a combination of ArUco, ARToolKitPlus and QR codes provides detailed information about the position of the unmanned vehicle in the warehouse, cargo identification and additional parameters, which together allows the successful use of the unmanned vehicle to move goods. The experiments were carried out on a JeVois-A33 smart machine vision camera with the JeVois Markers Combo software module. To configure the mode, the JeVois Inventor graphical interface was used. Experiments were carried out by scanning markers at right angles to the camera at a distance of 0.3 to 2 meters. Marker dictionaries with the smallest possible dimensions were used, namely 4 x 4 for ArUco and 3 x 3 for ARToolKitPlus. ArUco 4 x 4 dictionary values range from 0 to 49 (50 items) and ARToolKitPlus 3 x 3 dictionary values from 0 to 31 (32 items). 100 experiments were carried out for ArUco and 96 experiments for ARToolKitPlus. Based on the results of the experiments, the probabilities of correct recognition of each type of mark were determined. Possible errors in marker recognition and their impact on unmanned vehicles are analyzed, including the loss of additional motor resource, the loss of incorrect movement of cargo, as well as the loss of cargo due to incorrect reading of information. Based on the data obtained, a simulation model was created that takes into account all scenarios of correct and incorrect marker recognition. The model assumes a normal distribution of costs by vehicle resource, additional cargo movement and cargo cost. Simulations were carried out in Matlab for 100 000 experiments. Each scenario of marker recognition errors that could lead to losses was modeled separately. The average and total losses for each of the possible situations of recognition errors are determined. It has been established in which cases the greatest financial losses occur. The results of the work can be used to create and analyze the effectiveness of localization algorithms based on visual markers in a room.