Lesson 2 getting to know slam for the first time

ch02 First time to know SLAM

Classic vision SLAM frame

Vision SLAM The process includes the following steps :

Sensor information reading : In vision SLAM Mainly for the camera image information reading and preprocessing . If it's in a robot , There may also be a code disk 、 Reading and synchronization of information such as inertial sensors .

Visual Odometry (Visual Odometry,VO): The task of visual odometer is to estimate the motion of camera between adjacent images , And the local map .VO Also known as the front end (Front End).

The visual odometer will inevitably have cumulative drift (Accumulating Drift) problem .

Back end optimization (Optimization): The back end receives the camera pose measured by the visual odometer at different times , And loop back detection information , From noisy data , Optimize them , Get globally consistent trajectories and maps . Because of the connection VO after , Also known as the back end (Back End).

In the early stage EKF As a representative , Now take graph optimization as a representative .

In vision SLAM in , The front end is more relevant to the field of computer vision , For example, image feature extraction and matching , The back end is mainly filter and nonlinear optimization algorithm .

Loop back detection (Loop Closing): Loop detection determines whether the robot has reached the previous position . If a loopback is detected , It will provide information to the back end for processing .

Drawing (Mapping): It's based on the estimated trajectory , Create a map corresponding to the task requirements .

The form of map includes measurement map ( Accurately represent the positional relationship of map objects ) And topology map ( More emphasis on the relationship between map elements
system ) Two kinds of . There are also sparse maps and dense maps .

SLAM The mathematical representation of the problem

“ The radish carries the sensor and moves in the environment ”, Described by the following two things :

What is sports ? We should consider from k − 1 k-1k−1 Moment to k kk moment , Location of radish x xx How it changed .

Equation of motion :

What is observation ? Suppose the radish is in k kk Moment in x_k  A signpost was detected at y_j , We have to think about how this is described in mathematical language .

The observation equation :

These two equations describe the most basic SLAM problem : When you know the reading of motion measurement u , And sensor readings z when , How to solve the positioning problem ( It is estimated that x ) And drawing problems ( It is estimated that y)? At this time , We will take SLAM The problem is modeled as a State estimation problem : How to pass the measurement data with noise , Estimate internal 、 Hidden state variables ?

《 Vision SLAM Fourteen speak 》 Notes extract _ncepu_Chen The blog of -CSDN Blog