TITLE:
Inverse Detection for Deterministic Object Selectivity in Industrial Safety Monitoring
AUTHORS:
Gilles Verschueren, Robbe Noens, Ward Nica, Marc Juwet, Dino Accoto
KEYWORDS:
Inverse Detection, Deterministic Filtering, Object Selectivity, Industrial Safety Monitoring, Human-Robot Collaboration, Functional Safety
JOURNAL NAME:
Open Journal of Applied Sciences,
Vol.16 No.7,
July
29,
2026
ABSTRACT: Industrial safety sensors commonly detect object presence in monitored zones around hazardous machinery, but they generally do not identify whether the detected object is a human, product, fixed structure, or mobile system. Treating all detections conservatively is necessary for safety, but can cause unnecessary speed reductions or protective stops when known non-human objects frequently enter monitored zones. This paper proposes inverse detection, a deterministic object-filtering principle for improving selectivity in zone-based industrial safety monitoring. All detections are treated as human-equivalent by default. Detections may only be removed from the human-equivalent set when they can be verified as known non-human objects using explicit information such as geometry, pose, communication status, sensor consistency, and timeout conditions. Failed or incomplete verification returns the detection to the safety-zone evaluation. The framework consists of detection, filter-definition, verification, and safety-zone evaluation layers. Verification levels are defined for fixed structures, robot-manipulated objects, communicating mobile systems, and human-equivalent detections. A proof-of-concept implementation with multiple 2D LiDAR sensors shows that verified packages and mobile objects can be filtered, while humans, unknown objects, displaced objects, and communication failures remain active in the safety evaluation. The implementation is not a certified safety function, but demonstrates the feasibility of deterministic filtering as a transparent alternative to direct human classification.