Low-complexity camera ego-motion estimation algorithm for real time applications

TL;DR: This contribution presents a low-complexity camera egomotion estimation algorithm for real-time applications which uses a feature based approach for motion estimation and a new method is introduced for feature selection which limits the number of feature points to be tracked and has a low dependency on structure in the image.

Abstract: This contribution presents a low-complexity camera egomotion estimation algorithm for real-time applications. The algorithm uses a feature based approach for motion estimation. A new method is introduced for feature selection which limits the number of feature points to be tracked and has a low dependency on structure in the image. Both these factors are important in real time, applications, as lesser features to track result in lower computational complexity and lesser dependency on image structure results in smaller variations in computational time for different images. This gain in speed is achieved at the cost of a slightly reduced robustness and accuracy. This trade-off between speed and accuracy pays off particularly in static scenes where high reduction in computational cost is achieved without the accuracy penalty. This algorithm can be used in applications where an estimate of camera motion is required and low computational complexity is of primary concern.