In this study, we present an edge-based intelligent sensing system for access control that integrates real-time face recognition with an embedded suspicious-behavior detection module. Access control serves as the application scenario for this edge-based sensing framework, in which camera-captured observations are processed locally into identity- and behavior-level security decisions. The system performs dynamic user registration and classifies individuals as owners, guests, or strangers, each triggering a dedicated response. Two complementary strategies identify suspicious strangers: dwell-time accumulation near entrances and repeated-appearance monitoring, which flags an individual reappearing more than ten times within 10 min, capturing loiterers who evade dwell-time-only thresholds. The hardware platform is a low-cost Raspberry Pi, with a Python/OpenCV (Open Source Computer Vision Library) pipeline employing the Local Binary Pattern Histogram algorithm for face recognition. Experimental evaluation demonstrates recognition accuracies of 100, 95.56, and 95.23% for owners, guests, and strangers, respectively, with an average recognition latency of 0.39 s and a processing throughput of 12.87 frames per second. The suspicious-behavior detection module attains 95.10% accuracy with zero false alarms, confirming the feasibility of the proposed solution for edge-based intelligent security applications. The evaluation was conducted under controlled, single-entry laboratory conditions; future work will extend the system to multi-camera deployments and evaluate robustness under varying facial expressions, poses, and occlusions.