Edge-AI Based Smart Attendance and Behavioral Monitoring System: A Hybrid CNN-LSTM Approach for Resource-Constrained Environments

Description

Abstract—In developing nations such as Tanzania, the integration of technology in educational administration is
constrained by cost, intermittent connectivity, and limited technical support. Early low-cost attendance prototypes (e.g.,
colour-tag detection with photoresistors) are attractive but fragile: sensor readings drift under ambient-light variation and
rule-based timers cannot model real student behaviour. This paper proposes an Edge-AI upgrade, ‘Automatic Attendance
Taker 2.0’, which replaces analogue sensing with a hybrid computer vision and time-series pipeline that runs on
Raspberry Pi-class devices. We introduce Invariant Color-Tag Recognition (ICTR), a lightweight CNN with illumination
normalisation and augmentation to identify students robustly under daylight, shade, and indoor lighting. To monitor
behaviour, we model entry/exit logs as sequences and apply an LSTM-based anomaly detector that flags prolonged
absence and atypical movement patterns, triggering local alerts (SMS/WhatsApp gateway) without relying on cloud
services. A proof-of-concept evaluation on a pilot dataset with synthetic illumination augmentation demonstrates strong
identification accuracy and practical real-time latency on commodity edge hardware, suggesting a scalable pathway for
affordable school safeguarding and administration in resource-constrained environments.