Abstract:
This poster presents a parallel implementation of an object detection algorithm, as well as an improved pruning technique, which is an important part in an object detection implementation. We focus on face detection, even though the techniques developed are applicable to object detection in general. We implement the well known face detection algorithm of Viola-Jones and parallelize the important steps, and then come up with a better pruning algorithm when many nearby windows indicate a detection. We also consider the effect of multiple scales and present a pruning algorithm that minimizes the detected windows such that a single window indicating the presence of a detected face can be concluded. Our pruning algorithm maximizes the face detection such that a few false positives may be detected, but all faces present are correctly identified.
