Efficient filtering techniques for edge and motion detection.

摘要

Numerous algorithms have been developed which perform edge and motion detection on digital images. Some of the more robust algorithms make use of extensive two and three dimensional digital filters. The amount of computations required to perform these filtering operations in a straight forward manner is quite large. Unless a large amount of parallel computation is performed it would be impossible to implement some of these algorithms in real time. This dissertation presents three specific filtering techniques which can be used to reduce the computational complexity associated with the digital filtering required by some edge and motion detection algorithms. The techniques presented are as follows: hierarchical filtering to form a set of Gaussian filtered images; combination of Gaussian filtered images to approximate Gaussian derivative filtered images; and subsampling of Gaussian filtered images. These efficient filtering techniques are applied to a particular robust edge and motion detection algorithm developed here at the University of Virginia by Chengho Hsin. Additionally, several modifications to Hsin's algorithm are presented which simplify the required temporal filtering and velocity determination and realize a method for combining results from multiple spatial filtering channels. Results obtained using the efficient filtering methods developed here were found to be of equal quality to the results found using straight forward filtering techniques. However, The computational complexity associated with the implementation of Hsin's algorithm is reduced by approximately one order of magnitude.