Homography Estimation from Image Pairs with Hierarchical Convolutional Networks

Japkowicz, Nathalie; Nowruzi, Farzan Erlik; Laganière, Robert

doi:10.1109/iccvw.2017.111

Cited by 58 publications

(53 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As in [17,19], we are also using the COCO 2014 dataset (Microsoft Common Objects in Context) [22]. First, all the images are converted to gray-scale and are down-sampled to a resolution of 320 × 240.…”

Section: Datasetmentioning

confidence: 99%

STN-Homography: Direct Estimation of Homography Parameters for Image Pairs

Zhou

2019

Applied Sciences

View full text Add to dashboard Cite

In this paper, we introduce the STN-Homography model to directly estimate the homography matrix between image pair. Different most CNN-based homography estimation methods which use an alternative 4-point homography parameterization, we use prove that, after coordinate normalization, the variance of elements of coordinate normalized 3 × 3 homography matrix is very small and suitable to be regressed well with CNN. Based on proposed STN-Homography, we use a hierarchical architecture which stacks several STN-Homography models and successively reduce the estimation error. Effectiveness of the proposed method is shown through experiments on MSCOCO dataset, in which it significantly outperforms the state-of-the-art. The average processing time of our hierarchical STN-Homography with 1 stage is only 4.87 ms on the GPU, and the processing time for hierarchical STN-Homography with 3 stages is 17.85 ms. The code will soon be open sourced.Keywords Homography · STN · CNN Recently, some attempts have been made to tackle the homography estimation with CNN, and acquired higher accuracy than the ORB+RANSAC method. HomographyNet [17] defined the homography between two images by relocation of a set of 4 points, also known as 4-point homography parameterization. Their model is based on the VGG's architecture [18] with 8 convolutional layers, a pooling layer after every 2 convolutions, and 2 fully connected layers with an L2 loss

show abstract

Section: Datasetmentioning

confidence: 99%

STN-Homography: Direct Estimation of Homography Parameters for Image Pairs

Zhou

2019

Applied Sciences

View full text Add to dashboard Cite

show abstract

“…The model proposed in [20] is similar to the VGG (Visual Geometry Group) architecture [21] with eight convolutional layers, one max pooling layer after every two convolutional layers, two fully connected layers and an L2 loss function that is calculated from the square of the difference between the predicted and the ground truth four-point coordinate values. Nowruzi et al [22] proposed a hierarchical model that is stacked by the twin convolutional regression networks to estimate the homography between a pair of images, and improved the prediction accuracy of four-point homography compared with that of the work by DeTone et al [20].…”

Section: Related Workmentioning

confidence: 99%

“…DeTone et al and Nowruzi et al [20,22] focused on estimating the homography between pairs of image_patch_a and image_patch_b with centers being roughly aligned, which we call center-aligned image pairs. However, in our task, we need to estimate the homography between image_a and image_patch_b when their centers are not necessarily aligned, as shown in Figure 2.…”

Section: Related Workmentioning

confidence: 99%

“…The homography estimation of a center-aligned image pair is a special case of the homography estimation of a non-center-aligned image pair. Because deep networks require many data to train from scratch, we took a similar approach as that used by DeTone et al and Nowruzi et al [20,22] to generate a seemingly infinite dataset for the homography estimation of center-aligned and non-center-aligned image pairs from the Microsoft Common Objects in Context (MS-COCO) [23] dataset. The trained models were then directly used for the homography estimation of climbing robot wall maps.…”

Section: Related Workmentioning

confidence: 99%

“…Following other researchers [20,22], we also used the MS-COCO 2014 dataset. We selected 118,000 images from the trainval set of the MS-COCO 2014 dataset to prepare training samples and 10,000 images from the test set of the MS-COCO 2014 dataset to prepare test samples.…”

Section: Dataset Of the Center-aligned Image Pairmentioning

confidence: 99%

See 2 more Smart Citations

Deep Homography Estimation and Its Application to Wall Maps of Wall-Climbing Robots

Zhou

2019

Applied Sciences

View full text Add to dashboard Cite

When locating wall-climbing robots with vision-based methods, locating and controlling the wall-climbing robot in the pixel coordinate of the wall map is an effective alternative that eliminates the need to calibrate the internal and external parameters of the camera. The estimation accuracy of the homography matrix between the camera image and the wall map directly impacts the pixel positioning accuracy of the wall-climbing robot in the wall map. In this study, we focused on the homography estimation between the camera image and wall map. We proposed HomographyFpnNet and obtained a smaller homography estimation error for a center-aligned image pair compared with the state of the art. The proposed hierarchical HomographyFpnNet for a non-center-aligned image pair significantly outperforms the method based on artificially designed features + Random Sample Consensus. The experiments conducted with a trained three-stage hierarchical HomographyFpnNet model on wall images of climbing robots also achieved small mean corner pixel error and proved its potential for estimating the homography between the wall map and camera images. The three-stage hierarchical HomographyFpnNet model has an average processing time of 10.8 ms on a GPU. The real-time processing speed satisfies the requirements of wall-climbing robots.

show abstract

Deep Fundamental Matrix Estimation Without Correspondences

Poursaeed

Yang

Prakash

et al. 2019

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Estimating fundamental matrices is a classic problem in computer vision. Traditional methods rely heavily on the correctness of estimated key-point correspondences, which can be noisy and unreliable. As a result, it is difficult for these methods to handle image pairs with large occlusion or significantly different camera poses. In this paper, we propose novel neural network architectures to estimate fundamental matrices in an end-to-end manner without relying on point correspondences. New modules and layers are introduced in order to preserve mathematical properties of the fundamental matrix as a homogeneous rank-2 matrix with seven degrees of freedom. We analyze performance of the proposed models using various metrics on the KITTI dataset, and show that they achieve competitive performance with traditional methods without the need for extracting correspondences.

show abstract

Homography Estimation from Image Pairs with Hierarchical Convolutional Networks

Cited by 58 publications

References 17 publications

STN-Homography: Direct Estimation of Homography Parameters for Image Pairs

STN-Homography: Direct Estimation of Homography Parameters for Image Pairs

Deep Homography Estimation and Its Application to Wall Maps of Wall-Climbing Robots

Deep Fundamental Matrix Estimation Without Correspondences

Contact Info

Product

Resources

About