How Dual-Camera Proctoring Works (And Why One Camera Isn't Enough)

Online proctoring became a mainstream technology category in 2020, when much of the world moved exams from physical testing centers to home computers. In the years since, the technical approaches to ensuring exam integrity have evolved substantially — partly because cheating techniques have evolved alongside.

A pattern has emerged in industry research and proctoring-vendor analyses: single-camera proctoring, the dominant approach in early online exams, misses a meaningful fraction of cheating attempts that have become common since 2022. The architectural response — using two cameras simultaneously, typically a laptop webcam plus a paired mobile camera — is now considered the higher-integrity baseline for credentials that need to hold up to scrutiny.

This post explains how dual-camera proctoring works, what it catches that single-camera systems miss, and why the architectural choice has implications beyond exam integrity itself.

The single-camera limitation

A laptop webcam typically captures a roughly 60-degree field of view at a fixed distance from the candidate's face — usually 40 to 60 centimeters. This vantage point provides a clear view of the candidate's face, eyes, and the area immediately behind them.

What the laptop camera does not capture, by virtue of physics rather than software:

• The candidate's hands and desk surface. Most laptop cameras sit at the top of the screen, looking forward and slightly down at the face. The desk and hands are below the frame.
• The space behind the laptop. A candidate could have notes taped to the back of the screen, a phone propped on the keyboard, or a second person standing just out of frame.
• A second monitor or device. Many home workspaces include a second screen. A candidate could have an LLM session, course notes, or live communication with a remote collaborator on a screen the laptop camera cannot see.
• The room periphery. Anyone sitting beyond the laptop's narrow viewing cone — a friend coaching from across the room, for example — is invisible to the camera.

In testing-center environments, these blind spots were managed by a human proctor walking the room. In home environments, the human proctor was typically replaced by a single webcam, which inherits the limitations above.

Industry observations and academic research on remote-exam integrity (see, for example, work published through the International Center for Academic Integrity) have documented a corresponding rise in workarounds — cheating techniques specifically designed around what a single laptop camera can and cannot see.

The dual-camera architecture

Dual-camera proctoring addresses these limitations by adding a second camera with a different vantage point. In most current implementations:

• The first camera is the laptop's built-in webcam. It maintains the original role: watching the candidate's face, eye gaze, and immediate environment.
• The second camera is the candidate's mobile phone, paired with the exam session over WebRTC at the start of the exam. The phone is typically placed several feet behind the candidate, oriented to capture the desk, the candidate's hands, the screens visible to the candidate, and the broader room.

The two camera feeds are synchronized in time, so an AI analysis system can correlate events across both views. A candidate looking down at their lap, for example, can be cross-referenced with what the phone camera sees at the same moment — a phone? notes? a hand signal to someone off-camera?

This architecture closes most of the blind spots that single-camera systems leave open. It does not eliminate the possibility of cheating, but it raises the difficulty significantly.

What the AI actually flags

Modern proctoring AI systems analyze video feeds for patterns associated with integrity violations. The specific flag categories vary by platform, but a representative set used in production systems includes:

• No face detected. The candidate has stepped out of frame, covered the camera, or left the seat.
• Multiple people in frame. A second person has entered either camera's view.
• Phone visible. A mobile device is detected on the desk or in the candidate's hand (when the candidate's own phone is supposed to be the secondary camera, anomalies in its behavior are flagged).
• Notes visible. Printed materials are detected on the desk or being referenced.
• Looking away repeatedly. The candidate's gaze pattern shows sustained attention to something off-screen — a monitor, notes, or a person.
• Screen changes. A second monitor or unexpected window switch is detected via the laptop camera capturing reflections or behavioral cues.
• Audio anomalies. Voice activity detected during a quiet exam phase, suggesting coaching from off-camera.

Not every flag is decisive. A "looking away" event, for instance, might be a candidate consulting a sticky note (a violation) or stretching their neck (not a violation). The flag itself records the event; human review or a tiered scoring system determines how to weight it.

Why auditable AI grading matters

A consideration that is often missed in discussions of proctoring technology: the AI's grading decisions themselves need to be auditable. If a candidate disputes a flag — for example, the AI flagged "phone visible" when the object on the desk was actually a calculator approved by the exam — there needs to be a way to review the decision.

Modern proctoring platforms vary substantially in this respect. Some retain only the final integrity score and a list of flag types. Others, including Aveluate's implementation, retain the full decision trail: the model used, the prompt sent to it, the specific frame the flag references, the confidence score, and a timestamp.

The implication for credential trust is straightforward. A credential whose underlying grading decisions can be replayed end-to-end can withstand disputes. A credential whose grading is opaque cannot. Over time, the more auditable category tends to accumulate higher trust in the labor market, because employers and candidates alike can verify that the bar applied to one credential was applied to all credentials.

The privacy considerations

Proctoring inevitably involves observing candidates in their personal environments. Different platforms make different choices about what this looks like, and the choices have real implications for privacy and data protection.

A common architectural choice that significantly reduces privacy exposure is periodic snapshot capture rather than continuous video recording. Instead of recording 30+ minutes of continuous footage of the candidate's home, the system captures still frames at random intervals (typically every 1–3 seconds). Each frame is uploaded for AI analysis and retained according to a clearly-defined retention policy.

The privacy benefits of this approach include:

• Smaller dataset. A 30-minute exam produces roughly 50–75 MB of JPEG snapshots versus 180–500 MB of continuous video. Less data stored means less data exposed in the event of a security incident.
• No audio capture. Snapshot-based systems typically don't record audio, which substantially reduces the legal exposure under wiretapping and consent statutes that apply more strictly to audio than to video.
• No continuous video of the candidate's home. A still image of a desk and a face every few seconds is materially less invasive than 30 minutes of continuous footage.
• Bounded retention. Most credible platforms retain proctoring data for a defined window (commonly 90 days for unflagged sessions, longer for flagged ones), then purge it.

These choices don't eliminate privacy considerations entirely, but they represent a meaningful shift toward proportionate data collection — capturing what's needed for integrity verification without retaining more than necessary.

What this means for credential trust

The architectural depth of a proctoring system has direct implications for how much weight a credential carries in the labor market. The pattern that emerges from observation:

1. Credentials backed by deeper proctoring tend to retain value over time. As cheating workarounds become known, credentials that anticipated them remain trustworthy.
2. Credentials backed by lighter proctoring tend to lose value as workarounds spread. When a credential becomes known to be game-able, hiring managers stop weighting it.
3. Auditable grading provides resilience against disputes. Credentials that can show their work survive challenges; credentials that can't either accumulate dispute history or get quietly dropped from hiring evaluations.

For employers evaluating which proctored credentials to weight in hiring decisions, the technical depth of the proctoring matters as much as the presence of proctoring at all. For candidates choosing which credentials to invest in, the same logic applies in reverse: a credential issued by a platform with serious proctoring tends to do more work in a hiring funnel than a credential from a lighter platform, even when the badge labels look identical.

Where proctoring technology is heading

Several developments appear to be shaping the next generation of proctoring systems:

• Voice-presence detection without audio recording. Detecting that someone is speaking nearby — without storing what they're saying — addresses coaching detection while substantially reducing privacy exposure compared to full audio capture.
• Cross-session pattern analysis. Identifying candidates who exhibit similar suspicious patterns across multiple sessions, which can surface coordinated cheating networks that single-session analysis would miss.
• Reduced time-to-result. AI proctoring systems are converging on near-real-time analysis, which means flags can be reviewed and credentials issued faster after exam completion.

The broader trajectory is toward proctoring that is more rigorous in integrity but also more proportionate in privacy — capturing enough to verify what needs verifying, without retaining more than necessary.

Summary

Dual-camera proctoring exists because single-camera proctoring leaves blind spots that have become routinely exploited in remote-exam environments. The second camera, typically a paired mobile device, closes most of those blind spots by providing a complementary vantage on the desk, hands, and broader room.

The architectural choice has consequences beyond catching cheaters. It affects how much weight a credential carries in hiring, how disputes are resolved, and how privacy considerations are managed. For credentials that are intended to mean something durable in the labor market, dual-camera coverage paired with auditable AI grading is increasingly considered the baseline rather than an upgrade.

---

Aveluate uses dual-camera proctoring with auditable AI grading and snapshot-based capture for all verified credentials. Try a free demo to see how a proctored exam works, or read about the broader shift to verified credentials.