Prozessautomatisierung






  • Anmelden

Visuelle Odometrie und SLAM mit Stereo-Kameras

pointMatching.png

Estimating its ego-motion is one of the most important capabilities for an autonomous mobile platform. Without reliable ego-motion estimation no long-term navigation is possible. Besides odometry, inertial sensors, DGPS, laser range finders and so on, vision based algorithms can contribute a lot of information. Stereo odometry is a vision based motion estimation algorithm that estimates the ego-motion of a stereo camera through its environment by evaluating the captured images.

In robotics, the term odometry is usually used to describe the process of estimating the movement of a robot by measuring the rotation of its wheels. This is usually done by counting the increments delivered by a wheel encoder. Odometry in this sense is prone to errors, as wheels do always slip, so the measured movement will always be greater than the real one. The word odometry is composed from the Greek words hodos (meaning "travel", "journey") and metron (meaning "measure").

So in general we can say, odometry is the process of measuring or estimating the movement of an object by incrementally evaluating the data of one or several sensors that are attached to the object. We do not use the term odometry if we have an absolute measurement frame available (e.g. GPS). Odometry usually refers to relative measurements only.

Stereo Odometry - An Overview

stereoOdometrie.png

First of all, why do we need stereo odometry in robotics? Classical odometry that calculates the robot’s movements from counting the revolutions of the robot’s wheels is often deceived by wheel slippage, especially in outdoor terrain. GPS information may not always be available with the desired quality. Be it because the robot is operating in areas where GPS transmissions can not be received like in forests or mines or because the robot happens to be operating on a planet that has not yet been equipped with a set of GPS satellites, like on Mars.

Stereo odometry can determine the ego-motion of the stereo camera in all 6 degrees of freedom that are possible in a 3D world: 3 for translation and 3 for rotation. The process of stereo odometry follows a certain scheme:

  1. Acquire a pair of images from the left and right camera of our stereo rig.
  2. Find interest points (you may also call them features or landmarks if you prefer) in the images.
  3. Calculate the 3D coordinates of those interest points.
  4. Match the interest points between images taken from different viewpoints.
  5. Use the matches to calculate the motion (which means combined translation and rotation) between the two viewpoints.

A very comprehensive and excellent textbook regarding stereo vision is Hartley and Zisserman’s Multiple View Geometry in Computer Vision. Many basic but also many advanced techniques and algorithms are discussed there.

Publikationen zum Thema