Deep Hough-Transform Line Priors

Abstract

Classical work on line segment detection is knowledge-based; it uses carefully designed geometric priors using either image gradients, pixel groupings, or Hough transform variants. Instead, current deep learning methods do away with all prior knowledge and replace priors by training deep networks on large manually annotated datasets. Here, we reduce the dependency on labeled data by building on the classic knowledge-based priors while using deep networks to learn features. We add line priors through a trainable Hough transform block into a deep network. Hough transform provides the prior knowledge about global line parameterizations, while the convolutional layers can learn the local gradient-like line features. On the Wireframe (ShanghaiTech) and York Urban datasets we show that adding prior knowledge improves data efficiency as line priors no longer need to be learned from data.

A deep learning approach based on graphs to detect plantation lines

“…The results from our experiment demonstrated that the proposed approach is feasible to detect both plant and line positions with high accuracy. Moreover, the comparison of our method against [ 44 , 45 ] deep neural networks indicated that our method could return accurate results, better than those of the state-of-the-art, and, when compared against its baseline (Visual Features), an improvement from 0.907 to 0.951 occurred. As such, we intend to discuss here this improvement and the importance of graphs theory in conjunction with the DNN model.…”

“…The proposed method was compared with two recent state-of-the-art methods in Table 4 . Deep Hough Transform [ 44 ] integrated the classical Hough transform into deeply learned representations, obtaining promising results in line detection using public datasets. PPGNet [ 45 ] is like the proposed method since it models the problem as a graph.…”

“…To demonstrate this approach's effectiveness, experiments were conducted within a corn plantation field at different growth stages, where some plantation gaps were identified due to problems that occurred during the planting process. Moreover, to verify the robustness of our method with the addition of graphs, we compared it against both a baseline and other state-of-the-art deep neural networks, like [ 44 , 45 ]. Our study brings an innovative contribution related to extracting plantation lines under challenging conditions, which may support several precision agriculture-related practices, since identifying plantation lines in remote sensing images is necessary for automatic farming processes.…”

Line detection via a lightweight CNN with a Hough layer

“…Finally, in a very similar paper [19], authors use the sequence of Hough Transform, several convolutions and an inverse Hough Transforms as a part of their network for pixel-based segmentation of straight line segments. From the perspective of a general-purpose line detection, the dataset used to train and test the network -Wireframe [16] is inconsistent, as some clearly salient straight lines are absent from the desired answer (see images below).…”

“…From the perspective of a general-purpose line detection, the dataset used to train and test the network -Wireframe [16] is inconsistent, as some clearly salient straight lines are absent from the desired answer (see images below). We believe that the CNN proposed in [19] spends many parameters to distinguish between "important" and "unimportant" lines and, possibly, overfits to the dataset strongly.…”