An official website of the United States government
Measuring speed is a critical factor to reduce motion artifacts for dynamic scene capture. Phase-shifting methods have the advantage of providing high-accuracy and dense 3D point clouds, but the phase unwrapping process affects the measurement speed. This paper presents an absolute phase unwrapping method capable of using only three speckle-embedded phase-shifted patterns for high-speed three-dimensional (3D) shape measurement on a single-camera, single-projector structured light system. The proposed method obtains the wrapped phase of the object from the speckle-embedded three-step phase-shifted patterns. Next, it utilizes the Semi-Global Matching (SGM) algorithm to establish the coarse correspondence between the image of the object with the embedded speckle pattern and the pre-obtained image of a flat surface with the same embedded speckle pattern. Then, a computational framework uses the coarse correspondence information to determine the fringe order pixel by pixel. The experimental results demonstrated that the proposed method can achieve high-speed and high-quality 3D measurements of complex scenes.
more »
« less
Deep Convolutional Neural Network Phase Unwrapping for Fringe Projection 3D Imaging
Phase unwrapping is a very important step in fringe projection 3D imaging. In this paper, we propose a new neural network for accurate phase unwrapping to address the special needs in fringe projection 3D imaging. Instead of labeling the wrapped phase with integers directly, a two-step training process with the same network configuration is proposed. In the first step, the network (network I) is trained to label only four key features in the wrapped phase. In the second step, another network with same configuration (network II) is trained to label the wrapped phase segments. The advantages are that the dimension of the wrapped phase can be much larger from that of the training data, and the phase with serious Gaussian noise can be correctly unwrapped. We demonstrate the performance and key features of the neural network trained with the simulation data for the experimental data.
more »
« less
- Award ID(s):
- 1918260
- PAR ID:
- 10167916
- Date Published:
- Journal Name:
- Sensors
- Volume:
- 20
- Issue:
- 13
- ISSN:
- 1424-8220
- Page Range / eLocation ID:
- 3691
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
This paper presents an absolute phase unwrapping method for high-speed three-dimensional (3D) shape measurement. This method uses three phase-shifted patterns and one binary random pattern on a single-camera, single-projector structured light system. We calculate the wrapped phase from phase-shifted images and determine the coarse correspondence through the digital image correlation (DIC) between the captured binary random pattern of the object and the pre-captured binary random pattern of a flat surface. We then developed a computational framework to determine fringe order number pixel by pixel using the coarse correspondence information. Since only one additional pattern is used, the proposed method can be used for high-speed 3D shape measurement. Experimental results successfully demonstrated that the proposed method can achieve high-speed and high-quality measurement of complex scenes.more » « less
-
An increasingly popular machine learning paradigm is to pretrain a neural network (NN) on many tasks offline, then adapt it to downstream tasks, often by re-training only the last linear layer of the network. This approach yields strong downstream performance in a variety of contexts, demonstrating that multitask pretraining leads to effective feature learning. Although several recent theoretical studies have shown that shallow NNs learn meaningful features when either (i) they are trained on a single task or (ii) they are linear, very little is known about the closer-to-practice case of nonlinear NNs trained on multiple tasks. In this work, we present the first results proving that feature learning occurs during training with a nonlinear model on multiple tasks. Our key insight is that multi-task pretraining induces a pseudo-contrastive loss that favors representations that align points that typically have the same label across tasks. Using this observation, we show that when the tasks are binary classification tasks with labels depending on the projection of the data onto an r-dimensional subspace within the d k r-dimensional input space, a simple gradient-based multitask learning algorithm on a two-layer ReLU NN recovers this projection, allowing for generalization to downstream tasks with sample and neuron complexity independent of d. In contrast, we show that with high probability over the draw of a single task, training on this single task cannot guarantee to learn all r ground-truth features.more » « less
-
An increasingly popular machine learning paradigm is to pretrain a neural network (NN) on many tasks offline, then adapt it to downstream tasks, often by re-training only the last linear layer of the network. This approach yields strong downstream performance in a variety of contexts, demonstrating that multitask pretraining leads to effective feature learning. Although several recent theoretical studies have shown that shallow NNs learn meaningful features when either (i) they are trained on a single task or (ii) they are linear, very little is known about the closer-to-practice case of nonlinear NNs trained on multiple tasks. In this work, we present the first results proving that feature learning occurs during training with a nonlinear model on multiple tasks. Our key insight is that multi-task pretraining induces a pseudo-contrastive loss that favors representations that align points that typically have the same label across tasks. Using this observation, we show that when the tasks are binary classification tasks with labels depending on the projection of the data onto an π-dimensional subspace within the π β«π-dimensional input space, a simple gradient-based multitask learning algorithm on a two-layer ReLU NN recovers this projection, allowing for generalization to downstream tasks with sample and neuron complexity independent of π. In contrast, we show that with high probability over the draw of a single task, training on this single task cannot guarantee to learn all π ground-truth features.more » « less
-
To alleviate the cost of collecting and annotating large- scale point cloud datasets for 3D scene understanding tasks, we propose an unsupervised learning approach to learn features from unlabeled point cloud β3D objectβ dataset by using part contrasting and object clustering with deep graph neural networks (GNNs). In the contrast learn- ing step, all the samples in the 3D object dataset are cut into two parts and put into a βpartβ dataset. Then a contrast learning GNN (ContrastNet) is trained to verify whether two randomly sampled parts from the part dataset belong to the same object. In the cluster learning step, the trained ContrastNet is applied to all the samples in the original 3D object dataset to extract features, which are used to group the samples into clusters. Then another GNN for cluster- ing learning (ClusterNet) is trained to predict the cluster IDs of all the training samples. The contrasting learning forces the ContrastNet to learn high-level semantic features of objects but probably ignores low-level features, while the ClusterNet improves the quality of learned features by be- ing trained to discover objects that belong to the same se- mantic categories by using cluster IDs. We have conducted extensive experiments to evaluate the proposed framework on point cloud classification tasks. The proposed unsupervised learning approach obtained comparable performance to the state-of-the-art unsupervised learning methods that used much more complicated network structures. The code and an extended version of this work is publicly available via: https://github.com/lingzhang1/ContrastNetmore » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Journal Article:
- https://doi.org/10.3390/s20133691
