2021 IEEE International Conference on Signal Processing, Communications and Computing (ICSPCC) | 2021
Video Mosaic of Arbitrary Viewing Direction Oriented on DAS
Abstract
DAS (Distributed Apertures Sensors) system can acquire panoramic images from a distributed number of sensors, whose performance is mainly determined by the stitching algorithm. In DAS, the direction angle between sensors is large. Usually, rather than the whole panorama, vision with arbitrary viewing direction is required. In addition, low power consumption and high real-time stitching are expected in DAS. In this paper, we propose a stitching method for images with large capturing angle intervals, which can provide panorama with arbitrary observation direction. First, we project the captured image using cylinder model to deduce the feature difference between large capturing directions. Then the feature points are matched, projected back to original image, and screened out. Finally the distance parameter of stitching is estimated using the matched points, and the panorama of arbitrary observe direction is stitched. Additionally, we implemented the algorithm using FPGA, which is suitable for deployment in DAS systems with less computational resources. Experiment results suggest that the proposed algorithm improves the frame rate significantly (43.4fps) with no significant difference between the stitching effect of this paper and the comparison algorithm.