Sun Legong | Beijing University of Technology (original) (raw)

Papers by Sun Legong

Research paper thumbnail of Simultaneous Acquisition of High Quality RGB Image and Polarization Information using a Sparse Polarization Sensor

arXiv (Cornell University), Sep 26, 2022

Sparse polarization sensor + our compensation (a) Sensor array (b) RGB (c) DoLP (d) AoLP Figure 1... more Sparse polarization sensor + our compensation (a) Sensor array (b) RGB (c) DoLP (d) AoLP Figure 1. Compared to conventional polarization sensors, combining a sparse polarization sensor with a compensation method can enable more accurate RGB images and polarization information to be obtained. The outputs of the conventional polarization sensor (top) and our proposed sensor (bottom). The RGB and degree of linear polarization (DoLP) values indicate the peak signal-to-noise ratio (PSNR) (the larger the better) and the angle of linear polarization (AoLP) values indicate angular error (the smaller the better).

Research paper thumbnail of Accurate Polarimetric BRDF for Real Polarization Scene Rendering

Computer Vision – ECCV 2020, 2020

Polarization has been used to solve a lot of computer vision tasks such as Shape from Polarizatio... more Polarization has been used to solve a lot of computer vision tasks such as Shape from Polarization (SfP). But existing methods suffer from ambiguity problems of polarization. To overcome such problems, some research works have suggested to use Convolutional Neural Network (CNN). But acquiring large scale dataset with polarization information is a very difficult task. If there is an accurate model which can describe a complicated phenomenon of polarization, we can easily produce synthetic polarized images with various situations to train CNN. In this paper, we propose a new polarimetric BRDF (pBRDF) model. We prove its accuracy by fitting our model to measured data with variety of light and camera conditions. We render polarized images using this model and use them to estimate surface normal. Experiments show that the CNN trained by our polarized images has more accuracy than one trained by RGB only.

Research paper thumbnail of Degree-of-linear-polarization-based Color Constancy

2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

Research paper thumbnail of An Improved Normalized Cross Correlation Algorithm for Object Tracking

This work describes a novel algorithm for real time tracking of imaging targets in video sequence... more This work describes a novel algorithm for real time tracking of imaging targets in video sequence. Normalized Cross Correlation based on Box-Filtering (NCC-BF) is a widely used algorithm in practice. But NCC-BF still has lots of redundancies in exhaustive template matching process. Thus we put forward a sufficient termination condition based on an adaptive lower bound threshold function in this paper, and we have proved that if this termination condition is verified, the correlation score at current position must lower than the maximum correlation obtained in previous, thus the template can be proceed with next reference position with out executing the rest of operations in current position. So the redundancy in NCC-Based object tracking can be efficiently reduced by our novel termination condition described in this paper. The experimental results of our new algorithm and actual CPU time are reported.

Research paper thumbnail of Research on target detection algorithms in optical-electronic imaging

Several methods have been designed and developed for target detection. In order to verify the tar... more Several methods have been designed and developed for target detection. In order to verify the target detection algorithms for the photoelectric images, and compare the advantages and disadvantages between different algorithms, the target tracking algorithms verification platform was designed and implemented. This system consist of video source processing module, algorithm pretreating module, target segmentation module, target detection and target tracking module. Design the interface between modules for more implantable. Based on this system, target detection algorithms can be easily developed on PC without hardware support.

Research paper thumbnail of Perspective object tracking based on image feature extraction

In the course of tracking there are various transformations for the object, including flip, rotat... more In the course of tracking there are various transformations for the object, including flip, rotation, scaling, shear and so on. In this paper, we describe a new approach that can perform well in coping with object's perspective transformation. A transformation matrix which is modeled with 9-parameter matrix is employed to represent the deformation of the object. So as to simplify the calculation and increase the speed of the estimation for the perspective model, we choose feature points for calculation. In this paper, a new algorithm is proposed to remove the inaccurate matching points and calculate the final perspective parameters. The result indicates that the method of object tracking has high precision and fast speed.

Research paper thumbnail of Simultaneous Acquisition of High Quality RGB Image and Polarization Information using a Sparse Polarization Sensor

arXiv (Cornell University), Sep 26, 2022

Sparse polarization sensor + our compensation (a) Sensor array (b) RGB (c) DoLP (d) AoLP Figure 1... more Sparse polarization sensor + our compensation (a) Sensor array (b) RGB (c) DoLP (d) AoLP Figure 1. Compared to conventional polarization sensors, combining a sparse polarization sensor with a compensation method can enable more accurate RGB images and polarization information to be obtained. The outputs of the conventional polarization sensor (top) and our proposed sensor (bottom). The RGB and degree of linear polarization (DoLP) values indicate the peak signal-to-noise ratio (PSNR) (the larger the better) and the angle of linear polarization (AoLP) values indicate angular error (the smaller the better).

Research paper thumbnail of Accurate Polarimetric BRDF for Real Polarization Scene Rendering

Computer Vision – ECCV 2020, 2020

Polarization has been used to solve a lot of computer vision tasks such as Shape from Polarizatio... more Polarization has been used to solve a lot of computer vision tasks such as Shape from Polarization (SfP). But existing methods suffer from ambiguity problems of polarization. To overcome such problems, some research works have suggested to use Convolutional Neural Network (CNN). But acquiring large scale dataset with polarization information is a very difficult task. If there is an accurate model which can describe a complicated phenomenon of polarization, we can easily produce synthetic polarized images with various situations to train CNN. In this paper, we propose a new polarimetric BRDF (pBRDF) model. We prove its accuracy by fitting our model to measured data with variety of light and camera conditions. We render polarized images using this model and use them to estimate surface normal. Experiments show that the CNN trained by our polarized images has more accuracy than one trained by RGB only.

Research paper thumbnail of Degree-of-linear-polarization-based Color Constancy

2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

Research paper thumbnail of An Improved Normalized Cross Correlation Algorithm for Object Tracking

This work describes a novel algorithm for real time tracking of imaging targets in video sequence... more This work describes a novel algorithm for real time tracking of imaging targets in video sequence. Normalized Cross Correlation based on Box-Filtering (NCC-BF) is a widely used algorithm in practice. But NCC-BF still has lots of redundancies in exhaustive template matching process. Thus we put forward a sufficient termination condition based on an adaptive lower bound threshold function in this paper, and we have proved that if this termination condition is verified, the correlation score at current position must lower than the maximum correlation obtained in previous, thus the template can be proceed with next reference position with out executing the rest of operations in current position. So the redundancy in NCC-Based object tracking can be efficiently reduced by our novel termination condition described in this paper. The experimental results of our new algorithm and actual CPU time are reported.

Research paper thumbnail of Research on target detection algorithms in optical-electronic imaging

Several methods have been designed and developed for target detection. In order to verify the tar... more Several methods have been designed and developed for target detection. In order to verify the target detection algorithms for the photoelectric images, and compare the advantages and disadvantages between different algorithms, the target tracking algorithms verification platform was designed and implemented. This system consist of video source processing module, algorithm pretreating module, target segmentation module, target detection and target tracking module. Design the interface between modules for more implantable. Based on this system, target detection algorithms can be easily developed on PC without hardware support.

Research paper thumbnail of Perspective object tracking based on image feature extraction

In the course of tracking there are various transformations for the object, including flip, rotat... more In the course of tracking there are various transformations for the object, including flip, rotation, scaling, shear and so on. In this paper, we describe a new approach that can perform well in coping with object's perspective transformation. A transformation matrix which is modeled with 9-parameter matrix is employed to represent the deformation of the object. So as to simplify the calculation and increase the speed of the estimation for the perspective model, we choose feature points for calculation. In this paper, a new algorithm is proposed to remove the inaccurate matching points and calculate the final perspective parameters. The result indicates that the method of object tracking has high precision and fast speed.