Vision
Face landmarks
Measures eye aspect ratio, mouth movement, head position, and face presence.
Eye aspect ratioMouth aspect ratioHead offsetFace presence
Case study / Driver monitoring AI
Multi-Modal Driver Monitoring System
AI Driver Safety is an in-cabin computer vision system for driver drowsiness detection, distraction detection, and phone-use detection.
It turns video into annotated clips, event logs, safety scores, and reports for fleet safety, ADAS, telematics, and automotive AI teams.
PythonOpenCVMediaPipe Face LandmarkerONNX RuntimeYOLO11NumPyPydantic
Project
AI Driver Safety
Category
Driver monitoring AI, drowsiness detection, and in-cabin sensing
Core loop
Video frame, landmarks, object detection, event scoring, alerts, report
Signals
Eye closure, yawning, drowsiness, distraction, phone use, vehicle context, sensor drowsiness
Outputs
Annotated MP4, event JSON, CSV, summary JSON, HTML report, screenshots
Repository
github.com/Inferensys/ai-driver-safety
Detection scope
What the system tracks.
Drowsiness
Eye closure, yawning, and drowsy state events across video frames.
Distraction
Head position, face presence, visible phone use, and attention signals.
Review data
Timestamp, frame index, signal, score, severity, landmarks, boxes, and metadata.
Demo
Demo video.
A real cabin clip running through the AI driver monitoring pipeline.
Drowsiness detection
Distraction and phone-use detection
Annotated MP4, JSON/CSV, and HTML report
Repository
Review the AI Driver Safety codebase.
Python package, config profiles, demo assets, architecture notes, edge guide, tests, and sample outputs.
github.com/Inferensys/ai-driver-safetySource repositoryCore components
Driver monitoring AI components.
Computer vision, object detection, event rules, scoring, and exports.
Objects
Phone-use detection
Uses ONNX YOLO11 to identify visible cell phones.
ONNX RuntimeYOLO11Cell phoneDistraction
Signal logic
Event rules
Uses thresholds, frame counters, and cooldowns before recording an event.
Frame countersSmoothingThresholdsCooldowns
Scoring
Driver risk scoring
Combines fatigue, distraction, phone use, sensor, and vehicle signals.
Signal weightingSignal interaction rulesSeverityRisk score
Evidence
Report exports
Exports MP4, event JSON, CSV, summary JSON, and an HTML report.
MP4Events JSONCSVHTML report
Use cases
Use cases.
Fleet safety AI, ADAS driver monitoring, insurance telematics, and in-cabin sensing.
Fleet
Fleet safety monitoring
Detect drowsiness, distraction, and phone use across a trip.
Fleet safetyDriver coachingRisk reportsTrip review
ADAS
ADAS driver monitoring
Evaluate driver attention and readiness in assisted-driving prototypes.
ADASAttentionReadinessCabin camera
Telematics
Insurance telematics
Pair fatigue and distraction signals with harsh braking, lane drift, and speeding.
TelematicsHarsh brakingLane driftSpeeding
Fatigue
Driver fatigue detection
Track eye closure, yawning, drowsiness, and optional physiological signals.
FatigueEye closureYawningHeart-rate signal
Distraction
Phone-use and distraction detection
Detect visible phone use and head-position changes.
Phone useHead poseDistractionObject detection
Demo outputs
Screenshots.
Live monitor, event timeline, clip batch, and phone-use detection.
Live monitor
Live driver monitor.
Annotated frame, driver state, active events, and score.
Risk scoreDriver stateAnnotated frame
Event timeline
Event timeline.
Timestamped drowsiness, yawning, eye-closure, distraction, and phone-use events.
TimestampSignalSeverity
Batch summary
Batch summary.
Event counts, confidence, duration, and runtime details across demo clips.
Batch summaryEvent countsRuntime metrics
Phone use
Phone-use detection.
Visible cell phone detection recorded as a phone_use event.
Cell phoneONNXphone_use
Pipeline
From video frame to event log.
01
Read video
The runtime reads video or webcam frames and wraps each one as a typed FramePacket.
02
Detect signals
Face landmarks produce eye, mouth, head-position, and face-presence signals. ONNX adds phone detection.
03
Apply rules
Counters, thresholds, and cooldowns control event creation.
04
Score driver state
The scorer combines fatigue, distraction, sensor, and vehicle signals.
05
Export files
The system writes annotated media, DetectionEvent records, timeline data, summary JSON, CSV, and HTML reports.
Edge deployment
Designed for local video analysis.
Runs locally with OpenCV and MediaPipe. Supports ONNX Runtime for object detection.
Baseline
CPU profile
The edge CPU profile processes every second frame and writes output at 15 FPS for local video or webcam demos.
CPU15 FPS outputVideo filesWebcam
Face
MediaPipe model
Runs the face landmark model locally.
Face landmarksLocal modelOpenCVFallback
Objects
ONNX detector path
Runs YOLO11 phone detection through ONNX Runtime.
YOLO11ONNXPhone detectorEdge accelerators
Performance
Runtime metrics
Reports FPS, latency, runtime FPS, detector provider, and fusion model.
Source FPSLatencyRuntime FPSProviders
Tuning
Config profiles
Profiles cover default, edge CPU, night driving, and phone demo runs.
DefaultEdge CPUNight drivingPhone demo
FAQ
Driver monitoring AI FAQ.
Answers for teams evaluating drowsiness detection, distraction detection, phone-use detection, ADAS, and edge AI.
What is an AI driver monitoring system?
An AI driver monitoring system uses in-cabin sensors such as cameras and optional physiological or vehicle data to detect driver attention, fatigue, distraction, drowsiness, phone use, and readiness.
How does AI Driver Safety detect drowsiness?
AI Driver Safety reads face landmarks for eye closure, mouth movement, head position, and face presence. It checks those signals over time before recording drowsiness, yawning, eye-closure, or distraction events.
Does the system detect phone use while driving?
Yes. The phone-use demo uses an ONNX YOLO11 detector to identify visible cell phones, records phone_use events, and includes them in the same safety timeline.
Can this run on edge devices?
Yes. The project runs locally with OpenCV and MediaPipe on CPU, and it documents an ONNX Runtime path for object detection and edge accelerators.
Is this certified automotive safety software?
It is research and demo software, not certified automotive safety software. It is suited for prototypes, demos, and workflow design before a production safety program.
Output gallery
Demo artifacts.
Video, timeline, batch view, and phone-use detection screenshots.
Live monitor
Event timeline
Clip batch
Phone-use detection
Contact
Talk to the team about your AI system.
Share what you are building, where you need help, and what needs to ship next. We will reply with the right next step.
01
NDA available
We can start under NDA when the work requires it.
02
Direct team access
You speak directly with the team doing the technical work.
03
Clear next step
We reply with a practical recommendation on scope, implementation, or rollout.
30m
working session
Direct
team access