contributions:

• An improved tightly-coupled,optimization-based monocular visual-intertial state estimation framework with robust initialization,IMU bias calibration and loop closure
• Real-time implementation of state estimator and AR demo on mobile devices
• Open source release

system

system

chapter

• 1 IMU Pre-integration

imu pre-integration

• 2 LOOP DETECTION AND FEATURE RETRIEVAL

• LOOP DETECTION

  
  LOOP DETECTION
• FEATURE RETRIEVAL
  • 1 KLT tracker

  • 2 BoW Vector Matches

    
    feature retrieval

• 3 ESTIMATOR INITIALIZATION
  for :metric scale (feature depth), gravity vector, body velocity, and gyroscope bias.

• Vision-Only Structure from Motion

• Visual-Inertial Alignment(视觉和惯导对齐)
  we first recover the gyroscope bias, then initialize velocity, gravity vector, and met- ric scale.

  
  visual-inertial alignment