Human Detection from Omnidirectional Camera Using Feature Tracking and Motion Segmentation

摘要

This paper proposes a motion segmentation method on images which are captured by an omnidirectional camera. A simple unwrapping method is performed to convert an omnidirectional image into a panoramic image. Two consecutive panoramic images are used for motion analysis. Corner features are extracted from the image, and their locations are defined in local patches by dividing an image into grid cells. Then, each feature in previous frame is tracked to find its corresponding in the current frame. The affine transformation is performed using three corresponding features. The regions of moving object are detected as transformed objects which are different from the previously registered background. Morphological processing is applied for smoothing the motion region. Histogram vertical projection and boundary saliency are applied to segmenting the motion. Finally, the proposed motion segmentation method is used for human detection in omnidirectional images. The performance result shown the best detection rate is 97.25% at 0.3 false positive rate.