If one aims at describing the structure of unknown real-world signals, then a multi-scale representation of data is of crucial importance.
Lindeberg [194 ]
The meaning of real-time depends on the application, however, it usually implies that processing is performed quickly with respect to the dynamics of the system under consideration. When estimating motion for autonomous navigation, there are two constraints that define the meaning of 'real-time'. Both constraints derive from the need to accurately capture environmental dynamics. The first limit is related to the imaging system and the second is imposed by the navigation problem.
Temporal image sampling must occur sufficiently quickly to avoid temporal aliasing.
Motion processing must proceed with an update rate and latency that ensures navigation decisions are made using the true current state of the environment.
In this section it will be shown how scale space can be used to avoid temporal aliasing and that the inclusion of range information in scale space leads to a dynamic scale space which removes the overhead associated with more traditional approaches.
To avoid errors caused by temporal aliasing, it is important to consider the frequency domain characteristics of an image sequence (spatial and temporal sampling are considered). While we focus on image data here, this analysis applies equally to any spatiotemporal data source.
A camera's spatial sampling rate is defined by the...
TABLE OF CONTENTS 
