International Journal of Science, Engineering and TechnologyAuthors: Abhishek Choudhary, Sharukh, Sazid Ansari, Rohit Kumar, Assistant Professor Gautam Tyagi
Abstract: Traditional automated attendance monitoring environments suffer heavily from manual entry friction, administrative vulnerabilities like buddy-punching, and catastrophic data breaches due to central cloud server biometric processing. This paper introduces BioAttend, a novel, open-source architectural solution designed to overcome these bottlenecks by moving the complete lifecycle of biometric capture, validation, mapping, and authentication directly onto the browser-side runtime environment. Operating completely within a zero-server framework, BioAttend utilizes client-side TensorFlow.js via the face-api.js abstraction to parse real-time video feeds locally. The system executes three specific neural pathways: an optimized SSD MobileNet V1 framework for facial region localization, a FaceLandmarks68Net model to continuously isolate 68 coordinate vectors, and a FaceRecognitionNet architecture optimized for extracting deep 128-dimensional floating-point identification vectors. To defeat active spoofing vectors, we present an execution engine parsing structural shift in the Eye Aspect Ratio (EAR). This engine tracks physical user blinking signatures down to a specific temporal sequence, ensuring authentication occurs only upon confirmation of actual human physiological activity. Multi-factor safety checks are established using localized cryptographic handshakes integrated directly with a serverless Google Apps Script infrastructure to process transactional one-time passwords (OTPs) and send instantaneous administrative compliance logs. Computational profiles collected across heterogeneous operating systems demonstrate validation loops finishing inside 45ms with recognition accuracies of 99.4%, showing that robust institutional security, low latency, and full regulatory privacy compliance can be maintained concurrently within sandboxed device parameters.
