Due to the significance and complexity of rail surface quality inspections on production lines, manual completion is inefficient. Existing 3D vision techniques, such as stereo vision, are too complex and cumbersome fo...
详细信息
Localization and Navigation technology of mobile robots is becoming increasingly important in production and manufacturing factory. It is necessary to ensure localization and docking accuracy at high accuracy points a...
详细信息
作者：
Yang ZhangFan Xiao ShanGang HeSoftware Department
SHENZHEN GUANGWEI TECHNOLOGY CO LTD China and Software Department SHENZHEN GUANGWEI TECHNOLOGY CO LTD China Algorithm Department
SHENZHEN China and Algorithm Department SHENZHEN DELTRON INTELLIGEN TECHNOLOGY China Algorithm Department
SHENZHEN China and Algorithm Department SHENZHEN JINGYOU WISDOM EDUCATION CO.LTD China
Due to the randomness and non-periodic nature of the future posture of the human body, the prediction of the posture of the human body has always been a very challenging task. In the latest research, graph convolution...
详细信息
ISBN:
(纸本)9781450384155
Due to the randomness and non-periodic nature of the future posture of the human body, the prediction of the posture of the human body has always been a very challenging task. In the latest research, graph convolution is proved to be an effective method to capture the dynamic relationship between the human body posture joints, which is helpful for the human body posture prediction. Moreover, graph convolution can abstract the pose of the human body to obtain a multi-scale pose set. As the level of abstraction increases, the posture movement will become more stable. Although the average prediction accuracy has improved significantly in recent years, there is still much room for exploration in the application of graph convolution in pose prediction. In this work, we propose a new multi-scale feature suppression attention map convolutional network (AZY-GCN) for end-to-end human pose prediction tasks. We use GCN to extract features from the fine-grained scale to the coarse-grained scale and then from the coarse-grained scale to the fine-grained scale. Then we combine and decode the extracted features at each scale to obtain the residual between the input and the target pose. We also performed intermediate supervision on all predicted poses so that the network can learn more representative features. In addition, we also propose a new feature suppression attention module (FISA-block), which can effectively extract relevant information from neighboring nodes while suppressing poor GCN learning noise. Our proposed method was evaluated on the public data sets of Human3.6M and CMU Mocap. After a large number of experiments, it is shown that our method has achieved relatively advanced performance.
This report summarizes the fourth-place solution of the 'Vision Meets Algae' object detection challenge held on IEEE UV'2022 focuses on object detection in marine biology images obtained through the micros...
详细信息
Localization and mapping technology of mobile robots is important in production and warehousing. The changing warehousing environment always result in poor accuracy and robustness of the mobile robot's localizatio...
详细信息
Due to the significance and complexity of rail surface quality inspections on production lines, manual completion is inefficient. Existing 3D vision techniques, such as stereo vision, are too complex and cumbersome fo...
详细信息
ISBN:
(数字)9798350391916
ISBN:
(纸本)9798350391923
Due to the significance and complexity of rail surface quality inspections on production lines, manual completion is inefficient. Existing 3D vision techniques, such as stereo vision, are too complex and cumbersome for industrial use. This paper introduces a straightforward and fast line laser calibration method, along with a rail surface defect detection approach using line laser 3D reconstruction. By projecting line lasers onto the sides of a calibration board of specific thickness, the position of the laser centerlines in the side areas is extracted to ascertain the world coordinates of the corresponding point clouds. Four sets of line laser measurement devices are then employed to conduct a full coverage scan of the rail. A streamlined multi-camera joint calibration method is presented, which stitches together 3D point clouds from various perspectives to recreate the complete cross-sectional profile of the rail. After transforming into a surface depth variation map, defects are identified based on depth thresholds. The method not only precisely locates and categorizes defects but also reduces reconstruction errors by 0.1mm and simplifies the calibration process by eliminating the need for multiple rotations of calibration targets, thereby speeding up and enhancing the feasibility of industrial application. The approach overcomes the complexities and cumbersome processes associated with traditional methods not suited for industrial settings. The system can detect at a speed of 1.5m/s, identifying surface defects deeper than 0.3mm with a verified accuracy of less than 0.02mm, thus fulfilling the requirements for rail surface defect inspection.
Localization and Navigation technology of mobile robots is becoming increasingly important in production and manufacturing factory. It is necessary to ensure localization and docking accuracy at high accuracy points a...
Localization and Navigation technology of mobile robots is becoming increasingly important in production and manufacturing factory. It is necessary to ensure localization and docking accuracy at high accuracy points and routes. In this paper, we propose a high-accuracy localization method based on 2D laser, wheel odometry, and prior probability grid map. The algorithm can maintain the accuracy of robot trajectory estimation in an dynamic environment that changes over time. Two resolution prior probability grid maps are constructed to determine the initial pose of the mobile robot. A factor graph model is constructed, allowing multiple relative or absolute observations, to be added to the factor graph as factors. During localization, two local submaps are constructed in real time, and are jointly optimized with prior maps before being used to the local scan matching. The laser odometry is generated by using local submaps scan matching, and is added to the factor graph model along with the wheel odometry and prior map scan matching. At the same time, several artificial features are designed at high-accuracy docking target points, and ensure the final docking accuracy by identifying artificial features.
In preparation for observing holographic 3D content, acquiring a set of RGB color and depth map images per scene is necessary to generate computer-generated holograms (CGHs) when using the fast Fourier transform (FFT)...
详细信息
Speed estimate is a significant argument for motor control system. With the professional SoC, various speed algorithms for incremental encoders have been proposed. In this work, a new speed algorithm based on RISC-V D...
详细信息
This report summarizes the fourth-place solution of the “Vision Meets Algae” object detection challenge held on IEEE UV’2022 focuses on object detection in marine biology images obtained through the microscope. Fir...
This report summarizes the fourth-place solution of the “Vision Meets Algae” object detection challenge held on IEEE UV’2022 focuses on object detection in marine biology images obtained through the microscope. First, we experimented with a large number of backbones and necks to improve mAP by enhancing the model structure. Then, we designed and tested a variety of data augmentation schemes based on algal characteristics from a data perspective. Finally, with multiple models ensembled adopted, our methods achieve 57.579% mAP on the test set.
暂无评论