Dalian University of Technology

Peng Cheng Laboratory

Haojie Li

Zhongxuan Luo

Xin Fan

Wei Zhong

Zhihui Wang

Risheng Liu

Xinchen Ye

Lin Lin

Hong Zhang

Shengquan Li

Rui Xu

Boqian Liu

Shenglun Chen

Yuankai Xiang

The present invention discloses a disparity estimation optimization method based on upsampling and exact rematching, which conducts exact rematching within a small range in an optimized network, improves previous upsampling methods such as neighbor interpolation and bilinear interpolation for disparity maps or cost maps, and works out a propagation-based upsampling method by the way of network so that accurate disparity values can be better restored from disparity maps in the upsampling process.

Disparity estimation optimization method based on upsampling and exact rematching

A disparity image fusion method for multiband stereo cameras belongs to the field of image processing and computer vision. The method obtains pixel disparity confidence information by using the intermediate output of binocular disparity estimation. The confidence information can be used to judge the disparity credibility of the position and assist disparity fusion. The confidence acquisition process makes full use of the intermediate output of calculation, and can be conveniently embedded into the traditional disparity estimation process, with high calculation efficiency and simple and easy operation. In the disparity image fusion method for multiband stereo cameras proposed by the method, the disparity diagrams participating in the fusion are obtained according to the binocular images of the corresponding bands, which makes full use of the information of each band and simultaneously avoiding introducing uncertainty and errors.

Disparity image fusion method for multiband stereo cameras

The present invention belongs to the field of image processing and computer vision, and discloses an acceleration method of depth estimation for multiband stereo cameras. In the process of depth estimation, during binocular stereo matching in each band, through compression of matched images, on one hand, disparity equipotential errors caused by binocular image correction can be offset to make the matching more accurate, and on the other hand, calculation overhead is reduced. In addition, before cost aggregation, cost diagrams are transversely compressed and sparsely matched, thereby reducing the calculation overhead again. Disparity diagrams obtained under different modes are fused to obtain all-weather, more complete and more accurate depth information.

Acceleration method of depth estimation for multiband stereo cameras

The present invention discloses a large-field-angle image real-time stitching method based on calibration. First, a calibration algorithm is used to solve the positional relationship between cameras, and the prior information is used to solve a homography matrix between images. The system is easy to build, and the program is simple and easy to implement; an overlapping area ROI of images can be calculated by the homography matrix between images, and an energy model thereof can be built and solved with a graph cut algorithm; the graph cut algorithm has high time complexity and depends on the number of nodes in a graph; here, images are divided into layers, and solutions are obtained layer by layer and iterated; and finally, a stitched image is further optimized by simple linear fusion of stitching seams and histogram equalization of the stitched image.

Large-field-angle image real-time stitching method based on calibration

A multispectral camera dynamic stereo calibration algorithm is based on saliency features. The joint self-calibration method comprises the following steps: step 1: conducting de-distortion and binocular correction on an original image according to internal parameters and original external parameters of an infrared camera and a visible light camera. Step 2: Detecting the saliency of the infrared image and the visible light image respectively based on a histogram contrast method. Step 3: Extracting feature points on the infrared image and the visible light image. Step 4: Matching the feature points extracted in the previous step. Step 5: judging a feature point coverage area. Step 6: correcting the calibration result. The present invention solves the change of a positional relationship between an infrared camera and a visible light camera due to factors such as temperature, humidity and vibration.

Multispectral camera dynamic stereo calibration algorithm based on saliency features

The present invention discloses a multispectral stereo camera self-calibration algorithm based on track feature registration, and belongs to the field of image processing and computer vision. Optimal matching points are obtained by extracting and matching motion tracks of objects, and external parameters are corrected accordingly. Compared with an ordinary method, the present invention uses the tracks of moving objects as the features required for self-calibration. The advantage of using the tracks is good cross-modal robustness. In addition, direct matching of the tracks also saves the steps of extraction and matching the feature points, thereby achieving the advantages of simple operation and accurate results.

Multispectral stereo camera self-calibration algorithm based on track feature registration

The invention discloses a deep network lung texture recognition method combined with multi-scale attention, which belongs to the field of image processing and computer vision. In order to accurately recognize the typical texture of diffuse lung disease in computed tomography (CT) images of the lung, a unique attention mechanism module and multi-scale feature fusion module were designed to construct a deep convolutional neural network combing multi-scale and attention, which achieves high-precision automatic recognition of typical textures of diffuse lung diseases. In addition, the proposed network structure is clear, easy to construct, and easy to implement.

Deep network lung texture recogniton method combined with multi-scale attention

The invention discloses an unsupervised content-preserved domain adaptation method for multiple CT lung texture recognition, which belongs to the field of image processing and computer vision. This method enables the deep network model of lung texture recognition trained in advance on one type of CT data (on the source domain), when applied to another CT image (on the target domain), under the premise of only obtaining target domain CT image and not requiring manually label the typical lung texture, the adversarial learning mechanism and the specially designed content consistency network module can be used to fine-tune the deep network model to maintain high performance in lung texture recognition on the target domain. This method not only saves development labor and time costs, but also is easy to implement and has high practicability.

Unsupervised content-preserved domain adaptation method for multiple CT lung texture recognition

The present invention discloses a multispectral camera external parameter self-calibration algorithm based on edge features, and belongs to the field of image processing and computer vision. Because a visible light camera and an infrared camera belong to different modes, fewer satisfactory point pairs are obtained by directly extracting and matching feature points. In order to solve the problem, the method starts from the edge features, and finds an optimal corresponding position of an infrared image on a visible light image through edge extraction and matching. In this way, a search range is reduced and the number of the satisfactory matched point pairs is increased, thereby more effectively conducting joint self-calibration on the infrared camera and the visible light camera. The operation is simple and results are accurate.

Multispectral camera external parameter self-calibration algorithm based on edge features

The present invention discloses a disparity estimation method for weakly supervised trusted cost propagation, which utilizes a deep learning method to optimize the initial cost obtained by the traditional method. By combining and making full use of respective advantages, the problems of false matching and difficult matching of untextured regions in the traditional method are solved, and the method for weakly supervised trusted cost propagation avoids the problem of data label dependency of the deep learning method.

Disparity estimation method for weakly supervised trusted cost propagation

The present invention discloses a method for constructing a grid map by using a binocular stereo camera. A high-performance computing platform is constructed by using a binocular camera and a GPU, and a high-performance solving algorithm is constructed to obtain a high-quality grid map containing three-dimensional information. The system in the present invention is easy to construct, so the input data may be collected by using the binocular stereo camera; the program is simple and easy to implement. According to the present invention, the grid height is calculated by using spatial prior information and statistical knowledge, so that a three-dimensional result is more robust; and according to the present invention, the adaptive threshold of grids is solved by using spatial geometry, filtering and screening of the grids are completed, and thus the generalization ability and robustness of the algorithm are improved.

Method for constructing grid map by using binocular stereo camera

The present invention discloses a grid map obstacle detection method fusing probability and height information, and belongs to the field of image processing and computer vision. A high-performance computing platform is constructed by using a GPU, and a high-performance solving algorithm is constructed to obtain obstacle information in a map. The system is easy to construct, the program is simple, and is easy to implement. The positions of obstacles are acquired in a multi-layer grid map by fusing probability and height information, so the robustness is high and the precision is high.

Grid map obstacle detection method fusing probability and height information

The present invention provides a method for estimating high-quality depth map based on depth prediction and enhancement sub-networks, belonging to the technical field of image processing and computer vision. This method constructs depth prediction subnetwork to predict depth information from color image and uses depth enhancement subnetwork to obtain high-quality depth map by recovering the low-resolution depth map. It is easy to construct the system, and can obtain the high-quality depth map from the corresponding color image directly by the well-trained end to end network. The algorithm is easy to be implemented. It uses high-frequency component of color image to help to recover the lost depth boundaries information caused by down-sampling operators in depth prediction sub-network, and finally obtains high-quality and high-resolution depth maps. It uses spatial pyramid pooling structure to increase the accuracy of depth map prediction for multi-scale objects in the scene.

Method for estimating high-quality depth maps based on depth prediction and enhancement subnetworks

The present invention provides a method of dense 3D scene reconstruction based on monocular camera and belongs to the technical field of image processing and computer vision, which builds the reconstruction strategy with fusion of traditional geometry-based depth computation and convolutional neural network (CNN) based depth prediction, and formulates depth reconstruction model solved by efficient algorithm to obtain high-quality dense depth map. The system is easy to construct because of its low requirement for hardware resources and achieves dense reconstruction only depending on ubiquitous monocular cameras. Camera tracking of feature-based SLAM provides accurate pose estimation, while depth reconstruction model with fusion of sparse depth points and CNN-inferred depth achieves dense depth estimation and 3D scene reconstruction; The use of fast solver in depth reconstruction avoids solving inversion of large-scale sparse matrix, which improves running speed of the algorithm and ensures the real-time dense 3D scene reconstruction based on monocular camera.

Method for 3D scene dense reconstruction based on monocular visual slam

Provided is a method based on deep neural network to extract appearance and geometry features for pulmonary textures classification, which belongs to the technical fields of medical image processing and computer vision. Taking 217 pulmonary computed tomography images as original data, several groups of datasets are generated through a preprocessing procedure. Each group includes a CT image patch, a corresponding image patch containing geometry information and a ground-truth label. A dual-branch residual network is constructed, including two branches separately takes CT image patches and corresponding image patches containing geometry information as input. Appearance and geometry information of pulmonary textures are learnt by the dual-branch residual network, and then they are fused to achieve high accuracy for pulmonary texture classification. Besides, the proposed network architecture is clear, easy to be constructed and implemented.

Growing community of inventors

Liaoning, China

Haojie Li