How accurate is edge-based AI?

How accurate is edge-based AI?


How accurate is edge-based AI?

Using a dedicated ADAS camera, typical accuracies are as follows:

    • Lane Departure Warning  95%
    • Forward Collision Warning  95%
    • Pedestrian Detection Warning  90%

    Factors affecting ADAS accuracy include:

    • Placement and calibration of camera
    • Sensitivity settings in configuration
    • Very high speeds
    • Extreme weather
    • Poor light conditions
    • The need for the preceding vehicle to be in the same 'recognized lane'
    • Human-shaped standing signs or roadside dummies (however, portrait patterns on the preceding vehicle or on a roadside billboard rarely trigger)
    • Accurate inputs from the Left and Right indicators
    • Water stains, cement joints, or other line-like marks on the roadway.

     

    Using a dedicated ADAS camera, typical accuracies are as follows:

    • Fatigue                  95%
    • Distraction          95%
    • Phone                   95%
    • Smoking                               95%
    • Seat belt                               90%

    Factors affecting DSM accuracy include:

    • Placement and calibration of camera
    • Sensitivity settings in configuration
    • Frequently blinking
    • Deep sunken eyes or long eyelashes
    • Eye glass reflection
    • Sunglasses and glasses that block infrared
    • Masks, thick beards and other objects around the mouth
    • Strong backlighting
    • Seatbelt same color and texture as driver’s clothing
    • Similar shaped and sized objects to a cigarette, e.g. pens
    • Hand and cigarette must be in view at the same time
    • Hand must not cover more than 80% of the cigarette
    • Cigarette is directly facing the DSM camera (no discernible features)
    • Cigarette is held horizontally
    • Phone is held side-on to the camera (either vertically or horizontally)
    • At least 20% of the phone must be visible
    • Phone color same as background
    • Similar shaped and sized objects to a phone, e.g. walkie-talkies and e-cigarettes
    • Hands free calling not detected

    In many of the above cases, the AI system will typically trigger an alarm, preferring not to miss a genuine event even though it may be a false positive.

     

    Please also note that the DSM features on an AI Dashcam are largely dependent on the manufacturer firmware. While dashcams such as the VT3500-AI broadly support all main DSM features, its better detections are in the areas of Phone Use and Seatbelts. All other DSM features are better supported with a dedicated DSM camera.

     


      • Related Articles

      • What is the value of edge-based AI?

        What is the value of edge-based AI? Edge-based AI is AI logic that sits on the camera. It is designed to work in real-time to alert the driver to: Advanced Driver Assistance System (ADAS) events such as Forward Collision Warnings, Following Distance ...
      • SurfSight AI-MV Driver Facing Camera events

        The AI technology detects the persons smoking action (or movement of the upper body). The Artificial intelligence (AI) focuses on face and upper body movements to determine the type of distracted driving. And the Machine vision (MV) technology uses ...
      • Calibrate the AI-12 dashcam accelerometer

        Calibrate the AI-12 dashcam accelerometer The dashcam contains internal sensors that detect risky driving events. You need to calibrate the sensors before using the AI-12 dashcam. The AI-14 dashcam calibrates automatically. Note You can only ...
      • VisionTrack - Turn off in-cab alarms or deactivate an AI feature from a specific device

        To turn off in-cab alarms or deactivate an AI feature from a specific device: Platform Management Locate device Select the green arrow when the vehicle is ON Enter the ID & Password (admin//VT2000) Select Alarm Select AI App ADAS Tab for ADAS and ...
      • FAQ Overview

        Frequently Asked Questions I already have telematics - why do I need cameras? While telematics can offer a certain level of insight into incidents (such as location, speed, G-force etc.), cameras can offer a full 360-degree view around a vehicle in ...