PeriSafe: Perimeter Monitoring for Child and Cognitive-Support Safety
Disciplines
Robotics
Abstract (300 words maximum)
Unsupervised approach to road edges is a common precursor to pedestrian injury for children and cognitively impaired individuals. This project develops a perimeter-monitoring safety module that observes the area near curbs and driveways, recognizes approach-to-boundary intent, and triggers layered, non-contact interventions before a crossing occurs. The problem addressed is the lack of timely, context-aware warnings in residential and school-adjacent settings, where attention can lapse and caregivers may be briefly out of view. The proposed solution combines advanced vision with time-of-flight distance and thermal sensing, using deep-learning detection and multi-object tracking to identify people and estimate motion relative to a virtual “no-cross” line. Risk is assessed from approach speed, heading, and dwell near the boundary. When risk is elevated, the system first issues local audible/visual alerts; if approach continues, it commands a small quadruped to adopt a safe, low-speed, path-presence stance that discourages forward motion without physical contact, while preserving clear egress. Processing is performed on-edge to minimize latency and protect privacy; no facial identification is used, and video buffering is limited to short diagnostic windows. The feasibility scope includes curbside mock-ups, day/night calibration, and closed-loop tests of alerting and robot positioning on a marked test lane with varied approach angles and partial occlusions. Initial results seek to demonstrate dependable person detection and approach recognition, timely alerts issued before line crossing, consistent robot positioning that encourages pause/redirect behavior, and observable improvements in reaction time and route compliance versus a no-system baseline. Future development will expand sensing robustness (additional RGB-D or beacon inputs), add geofenced zones and caregiver notifications, introduce confidence-aware rechecks under occlusion or glare, and undertake formal safety, usability, and privacy evaluations to support deployment in residential neighborhoods, school drop-off zones, and community parks.
Use of AI Disclaimer
no
Academic department under which the project should be listed
SPCEET – Robotics and Mechatronics Engineering
Primary Investigator (PI) Name
Muhammad Hassan Tanveer
Additional Faculty
Fariha Alam, Mechatronics, falam5@students.kennesaw.edu
PeriSafe: Perimeter Monitoring for Child and Cognitive-Support Safety
Unsupervised approach to road edges is a common precursor to pedestrian injury for children and cognitively impaired individuals. This project develops a perimeter-monitoring safety module that observes the area near curbs and driveways, recognizes approach-to-boundary intent, and triggers layered, non-contact interventions before a crossing occurs. The problem addressed is the lack of timely, context-aware warnings in residential and school-adjacent settings, where attention can lapse and caregivers may be briefly out of view. The proposed solution combines advanced vision with time-of-flight distance and thermal sensing, using deep-learning detection and multi-object tracking to identify people and estimate motion relative to a virtual “no-cross” line. Risk is assessed from approach speed, heading, and dwell near the boundary. When risk is elevated, the system first issues local audible/visual alerts; if approach continues, it commands a small quadruped to adopt a safe, low-speed, path-presence stance that discourages forward motion without physical contact, while preserving clear egress. Processing is performed on-edge to minimize latency and protect privacy; no facial identification is used, and video buffering is limited to short diagnostic windows. The feasibility scope includes curbside mock-ups, day/night calibration, and closed-loop tests of alerting and robot positioning on a marked test lane with varied approach angles and partial occlusions. Initial results seek to demonstrate dependable person detection and approach recognition, timely alerts issued before line crossing, consistent robot positioning that encourages pause/redirect behavior, and observable improvements in reaction time and route compliance versus a no-system baseline. Future development will expand sensing robustness (additional RGB-D or beacon inputs), add geofenced zones and caregiver notifications, introduce confidence-aware rechecks under occlusion or glare, and undertake formal safety, usability, and privacy evaluations to support deployment in residential neighborhoods, school drop-off zones, and community parks.