Abstract—Ensuring that only the correct individual casts a vote has always been the fundamental challenge in any electoral system, and this challenge intensifies when voting is conducted online. Traditional e-voting systems have relied primarily on passwords, identification numbers, or one-time passcodes (OTPs) for authentication—methods that, while convenient, remain susceptible to credential theft and impersonation. This paper presents SecureVote+, a multi-layered secure online voting system that places the voter’s biometric identity at the center of the authentication pipeline. During voter registration, the system captures a live facial image and generates a unique 128-dimensional facial embedding using a deep convolutional neural network (CNN) based on the FaceNet dlib ResNet architecture for each user. At voting time, the system verifies the voter’s live face against the stored embedding using cosine similarity with a calibrated threshold of 0.82. Any attempt falling below the threshold is automatically rejected. The one person–one-vote enforcement mechanism prevents re-voting even when subsequent authentications exceed the threshold. Advanced Encryption Standard (AES) encryption secures all stored ballots, parameterized SQL queries on an SQLite 3 database protect the backend against injection attacks, and a blockchain-inspired SHA-256 audit chain records cryptographic hashes of all voting transactions to provide tamper-evident auditability. Experiments across 200 registered voter sessions organized in five progressive scenarios demonstrated an average facial authentication accuracy of 94.5%, with all duplicate voting attempts blocked and liveness detection operating consistently at 97.5% accuracy.