Embedded structured light cameras have been widely applied in various fields. However, due to constraints such as insufficient computing resources, it remains difficult to achieve high-speed structured light point cloud computation. To address this issue, this study proposes a memory-driven computational framework for accelerating point cloud computation. Specifically, the point cloud computation process is precomputed as much as possible and stored in memory in the form of parameters, thereby significantly reducing the computational load during actual point cloud computation. The framework is instantiated in two forms: a low-memory method that minimizes memory footprint at the expense of point cloud stability, and a high-memory method that preserves the nonlinear phase–distance relation via an extensive lookup table. Experimental evaluations demonstrate that the proposed methods achieve comparable accuracy to the conventional method while delivering substantial speedups, and data-format optimizations further reduce required bandwidth. This framework offers a generalizable paradigm for optimizing structured light pipelines, paving the way for enhanced real-time 3D sensing in embedded applications.
With the rapid development of information technologies such as 5G, big data, cloud computing, AIoT, and Geographic Information Systems (GIS), digital twin technology has received unprecedented attention and application as an effective approach to connect the physical and virtual worlds. Digital twin is essentially a digital model of a physical entity that maintains consistency with the physical counterpart. In order to achieve a precise mapping of the physical entity, a series of measures and algorithm models need to be adopted, and data assimilation provides an effective solution by integrating the observable states of the real-world object with the digital twin model to enhance the reliability of the twin model's output states. This article introduces the application methods and research status of data assimilation in digital twin, and makes preliminary discussions on the pressing issues that need to be addressed in the application of data assimilation methods in digital twin.
Deep learning-based industrial product surface defect detection methods are replacing manual inspection, while the issue of small object detection remains a key challenge in the current field of surface defect detection. The feature pyramid structures demonstrate great potential in improving the performance of small object detection and are one of the important current research directions. Nevertheless, traditional feature pyramid networks still suffer from problems such as imprecise focus on key features, insufficient feature discrimination capabilities, and weak correlations between features. To address these issues, this paper proposes a plug-and-play guided focus feature pyramid network, named GF-FPN. Built on the foundation of FPN, this network is designed with a bottom-up guided aggregation network (GFN): through a lightweight pyramidal attention module (LPAM), star operation, and residual connections, it establishes correlations between objects and local contextual information, as well as between shallow-level details and deep-level semantic features. This enables the feature pyramid network to focus on key features, enhance the ability to distinguish between objects and backgrounds, and thereby improve the model’s small object detection performance. Experimental results on the self-built TinyIndus dataset and NEU-DET demonstrate that the detection model based on GF-FPN exhibits more competitive advantages in object detection compared to existing models.
Camera calibration is the fundamental core link in the field of machine vision, and the calibration accuracy directly determines the upper limit of subsequent visual tasks. This paper proposes an automated high-precision internal parameter calibration method for cameras that combines adaptive exposure adjustment and multi pose collaboration to address the problems of traditional camera calibration processes being cumbersome, easily affected by environmental light fluctuations and the randomness of calibration plate pose adjustments, and poor stability of calibration results due to insufficient corner point extraction accuracy. This method first constructs a brightness evaluation standard based on the average grayscale value of the calibration image area, and designs an adaptive exposure adjustment algorithm with grayscale feedback compensation to stabilize the grayscale value of the calibration image in the optimal range; Secondly, the strategy of integrating Harris corner detection and grayscale centroid method is adopted to achieve high-precision extraction of calibration board corner points from pixel level coarse positioning to sub-pixel level refinement; Finally, combined with high-precision pose control of the robotic arm, a standardized calibration pose set is designed to achieve automatic acquisition of multi pose images of the calibration board. The experimental results show that the reprojection error of the proposed method is as low as 0.06 pixels, and the camera intrinsic parameter coefficient of variation is less than 6%. Compared with traditional calibration methods, this method has significant advantages in stability, robustness, and universality, and can provide reliable camera parameter support for industrial measurement, 3D reconstruction, and other fields.
Lane detection, as one of the key real-time tasks in vehicle tracking, often faces numerous challenges due to the complex shapes and inherent symmetry of lane lines. Aiming at the requirements of high accuracy and real-time performance of the lane detection, this paper proposes an adaptive structural optimization and enhanced receptive field network for lane detection (ASO-ERFNet). Specifically, we propose a Multi-receptive Field Fusion Layer (MFFL) that optimizes the structure of the pooling operation to fully utilize its hierarchical characteristics, leverage the symmetry of lane layouts, and precisely preserve the spatial geometric properties of lane lines. Then, we design a Dynamic Cross Attention Mechanism (DCAM) that can adaptively adjust convolutional kernel weights, thereby further enhancing detection precision. In addition, compared with the existing single-feature extraction networks, we propose a Dual-Perception Network (DPNet) that fuses features from different scale levels through a U-shaped structure. The experimental findings show that compared with state-of-the-art methods, our model obtains superior performance, achieving the best results in six scenes, such as normal and crowded scenes. Specifically, our method achieved 76.61% and 79.90% performance in crowded and shadow scenes, respectively, while reducing the false positive rate by 22.90% in cross scenes. Moreover, with an average processing time of only 5.5 ms per frame, our method achieves an F1 score of 96.65% on the TuSimple dataset and 76.23% on the CULane dataset. These results indicate that our model effectively balances detection accuracy and speed by leveraging lane symmetry, providing an efficient and reliable solution for lane detection in vehicle tracking.