Research of Fast Multi-scale Pyramid Spatio-temporal Interesting Points Extraction Algorithm

摘要

Extraction algorithm of multi-scale pyramid spatial-Temporal interesting points (STIPs) is widely applied to various vision applications, especially in pedestrian detection. The traditional algorithm's computational cost is too much to save much time, so a new method based on prediction is proposed in this paper. Firstly, put a radio into several groups on the basis of original pyramid. Then calculate the relationship between layer and layer. And finally, get low-level features of each adjacent layer by calculating the several layers'. In this way, when drawing the spatial-Temporal interesting points in the multi-scale Pyramid, a lot of time can be saved through this prediction method.