Wheel Odometry and Its Limitations #

“The simplest form of localization is to use wheel odometry methods that rely upon wheel encoders to measure the amount of rotation of robots wheels. In those methods, wheel rotation measurements are incrementally used in conjunction with the robot’s motion model to find the robot’s current location with respect to a global reference coordinate system. The wheel odometry method has some major limitations. Firstly, it is limited to wheeled ground vehicles and secondly, since the localization is incremental (based on the previous estimated location), measurement errors are accumulated over time and cause the estimated robot pose to drift from its actual location. There are a number of error sources in wheel odometry methods, the most significant being wheel slippage in uneven terrain or slippery floors.” ( Yousif et al., 2015, p. 289) ( pdf)

Visual Odometry (VO) #

“VO is the process of estimating the egomotion of an agent (e.g., vehicle, human, and robot) using only the input of a single or multiple cameras attached to it [101].” ( Yousif et al., 2015, p. 290) ( pdf)

“Whereas SLAM is a process in which a robot is required to localize itself in an unknown environment and build a map of this environment at the same time without any prior information with the aid of external sensors (or a single sensor). Although VO does not solve the drift problem, researchers have shown that VO methods perform significantly better than wheel odometry and dead reckoning techniques [54] while the cost of cameras is much lower compared to accurate IMUs and LASER scanners” ( Yousif et al., 2015, p. 290) ( pdf)

“VO is a particular case of a technique known as Structure From Motion (SFM) that tackles the problem of 3D reconstruction of both the structure of the environment and camera poses from sequentially ordered or unordered image sets” ( Yousif et al., 2015, p. 290) ( pdf)

“However, VO has been shown to produce localization estimates that are much more accurate and reliable over longer periods of time compared to wheel odometry [54]. VO is also not affected by wheel slippage usually caused by uneven terrain.” ( Yousif et al., 2015, p. 293) ( pdf)

“Full descriptions of different ways to solve the motion estimation using the above approach are provided by [75, 86, 119].” ( Yousif et al., 2015, p. 294) ( pdf)

Feature Extraction Techniques #

“Scale Invariant Feature Transform (SIFT) is recognized by many as one of the most robust feature extraction techniques currently available.” ( Yousif et al., 2015, p. 301) ( pdf)