ConductVision Enterprise
ConductVision counts teller and ATM queues, checks which teller windows are staffed and measures the line at opening for branch operations teams.
7 detectionsConductVision detections on this photo. Photo: Daivewoi Huongs, CC BY-SA 4.0, cropped.
Financial institutions invest heavily in physical security infrastructure, yet the vast majority of camera footage serves only a forensic purpose, reviewed after incidents have already occurred. ATM skimming, branch robberies, vault procedure violations, and cash-in-transit vulnerabilities persist because existing camera systems lack real-time analytical intelligence. ConductVision's technology can transform passive surveillance networks into active security platforms, providing real-time alerts for suspicious behavior at ATMs, branch security anomalies, cash handling procedure verification, and vault access monitoring. The result can be a shift from reactive footage review to proactive threat detection.
Outcomes
Teller windows open and close through the day as staff take breaks and cover other work.
Customers queue in front of windows with no one behind the counter.
ATM transaction logs record completed withdrawals and miss the people waiting in line.
Branches cannot tell when a second machine or longer lobby hours would help.
Customers line up outside before the branch opens on busy days.
A long first-hour line sets the wait for the rest of the morning.
ATM networks face evolving skimming threats, shoulder surfing, and physical attacks that are difficult to detect from transaction data alone. Branch locations must balance an open, accessible customer environment with security against robbery and internal fraud. Cash handling and vault operations follow strict procedures, but compliance verification depends on manual supervision that cannot cover all shifts and locations consistently.
ConductVision can process ATM camera feeds to detect skimming device installation, loitering, and shoulder surfing behavior in real time. Branch cameras can identify suspicious behavioral patterns and trigger silent alerts. Cash handling and vault operations can be monitored visually to verify procedural compliance, providing auditable records for internal controls.
What the model sees
ConductVision detections on this photo. Photo: othree, CC BY 2.0, cropped.
ConductVision detections on this photo. Photo: Canmenwalker, CC BY 4.0, cropped.
Endpoints
Each endpoint is defined before the pilot starts, with what is counted, its unit and how the results are broken down.
| Endpoint | How it is counted | Unit | Broken down by |
|---|---|---|---|
| PrimaryQueue per staffed window | People in the teller queue divided by staffed windows, sampled every 5 minutes. | Unit:people per window (hourly mean) | By:Branch, hour |
| SecondaryStaffed window rate | Share of open-hours minutes each teller window has a staff member behind the counter. | Unit:% | By:Window, hour |
| ATM line length | People waiting behind the person at each ATM, sampled every 5 minutes. | Unit:people (hourly peak) | By:ATM, hour, weekday |
| Opening line | People counted waiting at the entrance in the 10 minutes before opening. | Unit:people (peak) | By:Branch, weekday |
Each pilot measures ConductVision on your own footage against a sample checked by hand, before anyone relies on its numbers.
Queue count agreement
Difference between the ConductVision queue count and a reviewer's hand count on sampled frames.
people per frame (mean absolute difference) · by camera
Window status agreement
Share of sampled frames where the ConductVision staffed or empty call for a window matches a reviewer's.
% · by window, time of day
ATM person recall
Share of people a reviewer sees in sampled ATM frames that ConductVision also detected.
% · by atm, light conditions
Capabilities
Counts and tracks people in each camera view, by zone and hour.
Tracks posture, waiting and movement from pose keypoints in public spaces.
Flags people and vehicles inside the zones you draw on each camera.
Counts dense crowds and maps where people gather, section by section.
Use cases
Cameras at ATM vestibules can detect device tampering, card skimmer installation attempts, shoulder surfing proximity, and forced transaction scenarios. Alerts can be routed to security operations centers with annotated video clips and ATM location identification for rapid response.
Interior cameras can detect loitering near teller lines, concealment of identity, and behavioral patterns associated with robbery reconnaissance. During operating hours, real-time analytics can complement alarm systems by providing pre-incident situational awareness to security teams.
Cameras positioned at teller stations, cash counting rooms, and cash-in-transit loading bays can verify that dual-control procedures, counting protocols, and secure handoff sequences are followed correctly. Procedural deviations can be logged with timestamps for compliance reporting.
Cameras at vault doors and safe deposit areas can verify authorized access, monitor dual-custody requirements, and detect after-hours entry attempts. Visual access logs can supplement electronic audit trails for regulatory examinations.
Entry cameras can count customers by hour and day, measure wait times in queue, and track service completion rates at teller and advisor stations. Data can inform staffing models, branch layout optimization, and service level reporting for management review.
Related
Counts shoppers, queues and carts from the store's existing cameras.
Counts people and vehicles at public entrances and inspection lanes.
Counts people, vehicles and bikes at intersections, stops and public spaces.
Pilot
Tell us about your sites and cameras, and a ConductVision engineer replies with a scoping plan built around the endpoints above.
