ELSA AI Solutions

ELSA Solution

Four solutions that run on ELSA, the Physical AI Infrastructure OS. Applied directly to on-site equipment, they deliver smart mobility, escalator AI, a robot parking system and elevator remote care.

Escalator AI

AI video analytics detect and respond to risky passenger behavior and equipment anomalies in real time.

Escalator accidents are caused not only by the condition of the equipment but also by passenger behavior and on-site conditions. ELSA analyzes AI camera footage in real time on an edge server, detecting passenger posture, crowd density, direction of travel and abnormal movement.

It also analyzes equipment sensor data alongside the video, assessing passenger risk and equipment anomalies together and supporting a staged response according to the level of risk.

Components

A vision-based hazard recognition AI and a sensor-based anomaly detection AI work together.

AIoT-based escalator anomaly detection

Collects and analyzes sensor data from key escalator components to detect signs of abnormality and predict the likelihood of failure.

AIoT-based anomaly detection model

Drive motor vibration sensorAcceleration · vibration spectrum

Drive unit EOCRCurrent · overload signals

Time-series data collectionReal-time accumulation at the edge

Time-series analysis modelTransformer · Residual Block

Anomaly detection · predictive maintenanceFailure prediction · lifetime prediction

Detects equipment anomalies from the sensor data of key components.
Builds the AI model by fine-tuning a time-series foundation model on an escalator knowledge dataset.
Predicts the likelihood of failure and the remaining service life of key components to support predictive maintenance.

Vision-based accident and hazard recognition

Analyzes camera footage in real time to detect hazards arising from passenger movement and on-site conditions on the escalator.

Video-based accident and hazard recognition model

Depth CameraDepth video

CCTVRGB video

ELSA.Edge AI

Density MapDensity estimation

Velocity Extractor

FlowNetOptical flow · motion

Pose Extractor

Object DetectorObject detection

CLIP

JudgmentCongestionPostureObjects

AnnouncementsPassenger guidance

Alert propagationControl room · engineers

Management screenReal-time monitoring

Emergency stopStaged stop

Monitors escalator operation in real time through cameras.
Recognizes hazardous situations by analyzing video frame by frame.
Analyzes passenger speed and movement.
Detects risky behavior and abnormal movement by passengers.
Recognizes hazardous objects and restricted riders in real time and issues warnings.

AIoT escalator safety management dashboard — status, CCTV and control-panel events for four escalators at Gwangan Station, Busan Metro Line 2

Services

Vision-based AI analysisAnalyzes passenger posture, crowd density, direction of travel and abnormal movement in real time.
Detection of six major hazardsDetects major hazards including crowding, reverse riding, sudden starts and stops, fallen passengers, abnormal situations and wheelchair entry.
Staged hazard responseResponds in stages according to the level of risk: warning → announcement → inspection support → emergency stop.
Hazard analysis and history managementShows hazard occurrences and diagnostic results on the control screen and manages the history data.
Integrated equipment monitoringAnalyzes signals from key equipment such as the drive unit and step chain together with video data, managing passenger safety and equipment condition as one.

Deployed atBusan Transportation Corporation, Gwangan Station