Accurately localizing and estimating the 6D pose of the target installation slot is a core perception requirement for robot-assisted Curtain Wall Module (CWM) installation in high-rise construction. Current practice often relies on visual judgment and verbal coordination during alignment at height, which is time-consuming, error-prone, and safety-critical. This paper presents a lightweight, markerless, monocular vision pipeline for target-slot localization and 6D pose estimation. The proposed two-stage pipeline integrates a lightweight facade semantic segmentation network with geometric corner extraction and a Perspective-n-Point (PnP) solver. First, the segmentation model parses the facade into structural components, providing stable geometric cues. Second, a rule-based corner-extraction and ordering procedure constructs 2D-3D correspondences, which are fused with a predefined 3D CWM model via PnP to estimate the 6D pose of an installed reference panel. The target-slot pose is then obtained through deterministic pose propagation using the facade layout and known module spacing. In the MuJoCo simulation, across 36 target-slot cases, the proposed method achieves a median 3D translation error of 11.59 mm (mean 16.34 mm), a median planar error of 2.24 mm (mean 3.08 mm), and a median orientation error of 4.91 ∘ (mean 7.19 ∘ ). Compared with an AprilTag fiducial baseline, the method attains comparable in-plane target-slot localization and competitive orientation estimates while avoiding marker dependency.

Assistive computer vision technologies have the potential to significantly enhance workplace safety by enabling early detection of hazards and supporting proactive risk management. However, the development of such systems is constrained by the absence of comprehensive video datasets and clearly defined tasks that capture real-world hazard conditions. This study formulates pre-incident hazard recognition as a distinct assistive-vision problem, focusing on identifying unsafe states that precede incidents rather than the incidents themselves. To address this problem, we propose the Workplace Hazards Dataset (WHD), a balanced and diverse set of real-world videos representing five universal hazard categories in varied workplace settings. Furthermore, we establish a standardized benchmarking framework that evaluates state-of-the-art convolutional and transformer-based video models on both performance and inference-latency metrics to assess real-time feasibility. Experimental results show that the Multiscale Vision Transformer (MViT 16 × 4) achieves the highest accuracy (74.1%) while maintaining efficient inference speed, highlighting the importance of balancing recognition accuracy with processing time. Overall, this work defines a new benchmark task for assistive computer vision and provides the foundation for developing real-time hazard recognition systems that enhance safety and efficiency in high-risk environments. • Introducing Workplace Hazards Dataset (WHD), the first real-world hazard video dataset. • Benchmarking state-of-the-art CNN and transformer-based models for hazard recognition. • Evaluating dataset and model effectiveness in identifying real-world workplace hazards.

The installation of curtain wall modules (CWM) in high-rise buildings is a complex task that poses significant safety risks due to manual labor, especially when working at great heights. Traditional methods are labor-intensive, time-consuming, and expose workers to hazards such as falls and equipment malfunctions. To mitigate these risks and enhance operational efficiency, automation and precise positioning of CWMs are essential. Accurate detection of installation locations becomes critical, as it enables crane operators or autonomous robots to position CWMs safely and precisely. This study introduces a novel approach utilizing semantic segmentation for detecting CWM installation locations. To address the challenges of deploying deep learning models on edge devices in construction environments, we propose Lightweight Attention Network (LANet), a lightweight, single-stream semantic segmentation architecture. LANet incorporates an optimized transformer module for global context modeling with linear complexity, enabling efficient feature extraction while maintaining computational efficiency. Additionally, we have developed a custom curtain wall dataset tailored for automating CWM installation, which was used to train and evaluate LANet. Experimental results demonstrate that LANet achieves competitive segmentation accuracy with only 1.92 million parameters, delivering real-time performance at 262 FPS on an RTX 3090 GPU and 19 FPS on a standard Intel i7 CPU. These results make LANet highly suitable for deployment in resource-constrained environments.

Effective handling of construction and demolition waste (CDW) is crucial for sustainable resource recovery in industrial settings. This study introduces WasteXtract , a vision-based AI model for class-specific segmentation of CDW designed to automate waste monitoring and sorting at material recovery facilities (MRFs). The theoretical contribution of this research lies in adapting large-scale vision foundation models by incorporating a class-specific mask decoder and a lightweight adapter in a ViT-based image encoder. This approach enables efficient and scalable fine-tuning for the industrial application of CDW recognition, achieving reliable waste detection and classification performance. The engineering applications of WasteXtract are validated across two datasets representing real-world recognition of waste composition in skip-bins and conveyor-belt scenarios for intelligent automation and optimised resource recovery. The segmentation results show that WasteXtract achieves mean weighted intersection over union (WIoU) scores of 0.66 in skip-bin waste composition monitoring and 0.96 in the conveyor-belt scenario. Compared to baseline models such as DeepLabv3+ and U-Net, WasteXtract achieves significantly higher segmentation accuracy while reducing the number of trainable parameters, demonstrating its suitability for deployment in real-world, resource-constrained environments such as MRFs. This study highlights the transformative role of informatics in advancing automation and efficiency within sustainable waste management.