Dyah Lestari, Siti Sendari, Wahyu Sakti Gunawan Irianto, Anik Nur Handayani, Samsul Setumin
Autonomous mobile robots often struggle to navigate dynamic environments where obstacles may appear, disappear, or move unpredictably. Traditional static mapping approaches fail to provide real-time environmental awareness, particularly in grid-based navigation systems. This research presents a vision-based Bayesian Occupancy Grid Mapping approach for detecting and localising environment changes in dynamic indoor environments. A top-down RGB camera is used to observe a grid workspace, where each object - robot, target, static obstacles, and a dynamic obstacle-is visually distinguished using Hue, Saturation, Lightness colour thresholding. The system maps each detected object to a specific grid cell index and continuously updates occupancy probabilities using Bayesian inference. Changes in the robot environment, such as object movement, appearance, and disappearance, are detected by comparing the current occupancy state with the previous one. The proposed method achieves detection and mapping accuracy of 93.6%, with a precision of 100%. Change detection can be performed precisely for object addition, removal, and displacement. The processing time from object detection to obtaining the changes of the robot's working environment averaged about 31 milliseconds per frame. Despite limitations related to lighting sensitivity, this system provides a low-cost and real-time mapping solution that is suitable for low-speed robot navigation and path planning in dynamic environments. © 2025 IEEE.
Universitas Negeri Malang, Department of Electrical Engineering and Informatics, Malang, Indonesia; Universiti Teknologi Mara, Cawangan Pulau Pinang, Centre for Electrical Engineering Studies, Permatang Pauh Campus, Pulau Pinang, Malaysia
