The Method for Modeling Three-Dimensional Change Event Modeling Method in Industrial Warehouses in a ROS 2/Gazebo Simulation Environment
Year: 2026
Type: article
Abstract: Repeated and fragmented point-cloud responses complicate the interpretation of warehouse changes, particularly when objects appear and disappear after baseline mapping. This study implemented lifecycle-aware voxel memory and event aggregation in ROS 2/Gazebo. Occupancy introduced after baseline construction was retained for subsequent removal decisions. Filtering, spatiotemporal clustering, and duplicate suppression generated event candidates, which were evaluated through one-to-one matching. The main LiDAR experiment aggregated 29,534 raw records into 11 event candidates, with 6 true positives, 5 false positives, and an F1 score of 0.706. Across 30 shared-input supplementary simulation sequences, the baseline-only and stale-state and radius-constrained memory variants achieved the mean recall values of 0.644 and 0.900, respectively. Their precision values were 0.662 and 0.532, and their F1 scores were 0.651 and 0.667. The 50 archived runs per sensor configuration yielded a mean fusion-minus-LiDAR F1 difference of 0.0099, with an exploratory 95% bootstrap interval of [−0.0204, 0.0395]. An exploratory geometric motion filter was also evaluated, with mask agreement reported separately from event scores. The experiments support simulation feasibility under model-state-assisted moving-actor masking, predefined regions of interest, and offline event processing.
Source: Applied Sciences
DOI: 10.3390/app162211187
Institution: South Ural State University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,896
OpenAlex ID: W7225872539
