Feature matching is a crucial technique to estimate robot pose in the light of natural landmarks for a vision navigation system. It is difficult for the present feature matching approaches to balance the accuracy, robustness and efficiency for a vision-based robot in complex workspaces. A multi-stage KNN-TF-BF-EN-RANSAC (KTBER) refinement matching technique is proposed to combine with the adaptive ORB feature extraction method, in order to develop a novel KTBER_AORB model that can outperform the state-of-the-art feature extraction and matching techniques. Firstly, the adaptive ORB (AORB) features are extracted in the Laplacian Transformation of Gaussian (LTOG) pyramid by using a set of variable-radius templates. The template radius is adjusted in terms of different brightness and contrast conditions. Then, the Eigenvector Norm (EN) and Bidirectional Filtering (BF) techniques are devised to establish a novel KTBER model, which can match AORB feature pairs not only according to the local appearance-based similarity of feature descriptors in the Hamming space, but also by means of the similarity of local image structure and global grayscale information for feature pairs in the Euclidean space. Finally, a large number of feature matching experiments are conducted on five public datasets and our mobile robot dataset, in order to test the algorithm performances for different challenging scenarios, including illumination change, motion blur, image rotation, viewpoint change, and low texture. On the one hand, an intensive experimental analysis of our model with different configurations is made to check the specific effects of different algorithm elements. On the other hand, the KTBER_AORB model is further compared with four state-of-the-art techniques, e.g., SIFT, ORB-SLAM, MP-ORB, and LAAT. The experimental results demonstrate that the KTBER_AORB model outperforms other state-of-the-art feature matching techniques on accuracy and robustness, still preserving a higher efficiency than most counterparts.

Nowadays, pointer instruments remain the main state monitoring devices in the power industry, because they have strong mechanical stability to resist electromagnetic interferences compared with digital instruments. Although the object detection algorithms based on deep learning have widely been used in the field of instrument detection, the meter recognition process still relies on threshold segmentation to recognize object points and on Hough transform to extract the meter pointer. An intelligent vision recognition method based on YOLOv5 and U²-Net network (YLU²-Net) is proposed to improve the accuracy and efficiency of meter recognition in a complex environment. Firstly, the pointer meter is located in the instrument images by using the YOLOv5 network as a region of interest (RoI). Then, the instrument RoI is processed by means of perspective transformation and image resizing. Thirdly, an improved U²-Net image segmentation method with the deep separable convolution and the focal loss function is devised to distinguish the pointers and scales from the background in the instrument RoI. Further, a dimension reduction reading method with the polar coordinate transformation is developed to calculate the meter reading accurately and efficiently. Finally, the ablation experiment is conducted to test the performance of each algorithm module in our method, and the competition experiment is completed to compare our method with other state-of-the-art ones. The experimental results verify the accuracy and efficiency of the YLU²-Net recognition method proposed.

Pose estimation is a crucial technique to achieve robot localization using natural features for an RGB-D vision navigation system. However, it still suffers from the impoverishment of robustness and accuracy in complex workspaces. A robust real-time pose estimation approach using hybrid point and twin line reprojection is proposed to allow accurate localization for a vision-based robot under various challenging scenarios, including low texture, low structure, occlusion, and viewpoint and illumination changes. First, an improved adaptive threshold (IAT)-based oriented fast and rotated brief (ORB) feature detection method is developed to extract sufficient IAT-ORB point features that are matched by a multistage ${K}$ -nearest neighbor (KNN)-ratio-test criterion (RTC)-cosine similarity (CS)-random sample consensus (RANSAC) (KRCR) refinement feature matching strategy to generate reliable point matches. Then, a robust EDLines line feature matching method is developed by integrating dynamic matching (DM) and median absolute deviation (MAD) to obtain fine line matches. Furthermore, the hybrid point reprojection errors are built from the fusion of 3-D–2-D and 3-D–3-D point pairs, and the twin line reprojection errors are constructed by integrating virtual right-eye lines generated by depth measurement, which both are incorporated into a novel unified errors’ optimization model to achieve optimal motion estimation by minimizing the reprojection errors of points and lines concurrently. Finally, the extensive experiments verify the robustness, accuracy, and real-time performance of the proposed approach on the public datasets and our real indoor scenarios. The quantitative results demonstrate that our proposed approach gains more accurate and stable localization than other state-of-the-art methods under challenging environments.