The use of feature detection is a standard process within the computer vision community for simplifying a complex image to a more manageable representation. Feature detection is typically applied to individual images, even if they are a part of a more extensive image sequence. In this thesis we present new methods for feature detection via phase congruency, applied to image sequences. This work shows the improvements that can be gained from taking an image in its context. The first section of work focuses upon extending a previous feature detector, phase congruency, to operate on an image sequence. This new technique shows improvements in the robustness of the feature detector under increasing levels of noise. It also improves feature orientation description allowing for the component velocity of a feature to be evaluated. After further evaluation however this method produced undesirable results for fast moving features. In response to this, a novel method for evaluating phase congruency has been developed. The new method is achieved by modelling the filtering process used to derive phase congruency by measuring the standard deviation of the normalised energy response. Accordingly, the new method is termed statistical phase congruency. This new approach is implemented first for 2-D images, showing improvements over the initial image-based phase congruency technique. Furthermore, it is extended to detect time persistent features in image sequences whilst also providing improved results for detecting fast moving features. It is intended that the results of this work will provide a basis for detecting time persistent features under noisy conditions. The final portion of this thesis gives some conclusions and adds some direction for future work on these ideas.