Name
Abstract | ||
|---|---|---|
With the growing use of digital lightweight cameras, generating 3D information has become an important challenge in computer vision. Despite several attempts presented in the literature to solve this challenge, it remains an open problem when it comes to the structural accuracy of the depth map and the required baseline (distance between the first and the last frames) to capture a sequence of images. In this paper, a novel approach is proposed to compute a high quality dense depth map together with a semi-dense/dense 3D structure from a sequence of images captured on a narrow baseline. Computing the depth information from small motions has been a challenge for decades because of the uncertain calculation of depth values when using a small baseline - up to 12mm. The proposed method can, in fact, perform on a much wider range of baselines from 8 mm up to 400 mm while respecting the structure of the reference frame. The evaluation has been done on more than 10 sets of recorded small motion clips and for the wider baseline, on 7 sets of stereo images from Middlebury benchmark. Preliminary results indicate that the proposed method has a better performance in terms of structural accuracy in comparison with the current state of the art methods. Also, the performance of the proposed method remains stable even when only a low number of frames are available for processing. |
Year | Venue | Field |
|---|---|---|
2017 | ICCV Workshops | Reference frame,Computer vision,Open problem,Pattern recognition,Computer science,Baseline (configuration management),Feature extraction,Artificial intelligence,Depth map |
DocType | Citations | PageRank |
Conference | 0 | 0.34 |
References | Authors | |
0 | 2 | |
Name | Order | Citations | PageRank |
|---|---|---|---|
| P. M. Corcoran | 1 | 414 | 82.56 |
| Hossein Javidnia | 2 | 10 | 4.71 |