多使用者混合實境工程機具虛驚事故應變系統

Abstract

營造業作為全球最具危險性的產業之一，工安意外與職業災害的發生率始終居高不下。在各種主動式的安全策略中，虛驚事故（near-miss）的偵測與分析因其能提供預警並避免意外發生而日益受到重視。然而，傳統的安全管理方法往往無法即時掌握現場情況並提供及時干預措施，例如透過人工監督及事件後的事故報告。近年來，受惠於數位科技的快速進展，透過實現即時感測與沉浸式的可視化技術為這些傳統方法的限制帶來了新契機。儘管如此，許多現有的解決方案在以工人為中心導向的監控方面仍然面臨挑戰，主要原因包括:視線盲點及攝影機覆蓋程度導致觀察不完整，抑或是因為標示未能正確辨識或視覺線索被忽略而導致資訊不充足。混合實境（Mixed Reality, MR）技術為此提供了可行的解決方法，透過沉浸式與空間精準的可視化來強化風險識別，並輔助即時決策與促進協同安全管理。

本研究提出一套支援多使用者的MR虛驚事故回應系統，該系統用於提升虛驚事故的危害偵測，並在施工現場提供及時警示。本研究所提出之系統整合 MR 與建築資訊模型（Building Information Modeling, BIM）資料、實境擷取資料（點雲）及固定式攝影機網絡，用於偵測接近中的施工設備並對 MR 使用者發出及時警示。警示訊息透過多模態回饋傳遞，包括視覺警示標誌、三維方向指示器與聲音警報。系統亦支援多用戶之間的溝通協作，並允許使用者間共享危險警示與協同應對之決策。

本系統依據四階段框架進行開發：(1) 資料註冊、(2) 虛擬場景生成、(3) 物件偵測以及 (4) 基於MR的虛驚事故警示可視化。物件偵測採用YOLOv8模型執行，MR裝置則是使用Microsoft HoloLens 2。偵測結果透過Wi-Fi網路上的UDP連線傳送至HoloLens，虛驚事故事件則是直接在HoloLens根據位置的分析來進行識別。本研究透過兩組實驗場景進一步驗證，系統偵測的準確率達到91.3\%，並展現平均180毫秒的即時反應效能。此外，多用戶通訊模組的延遲僅為103毫秒。研究亦透過九位建築從業人員參與的半結構式訪談回饋進一步證實該系統在強化情境感知、支援使用者應對及提升協同安全管理方面具備實務可行性與有效性。

The construction industry remains one of the most hazardous sectors worldwide, consistently experiencing high rates of occupational accidents. Among various proactive safety strategies, the detection and analysis of near-miss events have become increasingly important for providing early warnings and preventing future accidents. However, traditional safety management practices, such as manual supervision and retrospective incident reporting, often fall short in delivering real-time situational awareness and timely intervention. Recent advances in digital technologies, especially those enabling real-time sensing and immersive visualization, offer new opportunities to address these limitations. Nevertheless, many existing solutions struggle with worker-centered monitoring due to incomplete observation, caused by blind spots and limited camera coverage, and insufficient information, such as unrecognized signage or overlooked visual cues. Mixed Reality (MR) technology presents a promising approach to overcome these challenges by enabling immersive, spatially accurate visualizations that enhance hazard recognition, support real-time decision-making, and promote collaborative safety management.

This study introduces a multi-user MR near-miss response system that enhances near-miss hazard detection and alert delivery in construction environments in real-time. The proposed system integrates MR with Building Information Modeling (BIM) data, reality capture data (point clouds), and a fixed camera network to detect approaching construction equipment and issue timely warnings to MR users. Alerts are delivered through multimodal feedback, including visual warning signs, 3D directional indicators, and auditory alarms. The system also facilitates multi-user communication, allowing for shared hazard awareness and coordinated decision-making.

The system was developed following a four-stage framework: (1) data registration, (2) virtual scene generation, (3) object detection, and (4) MR-based near-miss alert visualization. Object detection is performed using the YOLOv8 model, and the Microsoft HoloLens 2 serves as the MR device. Detection results are transmitted to the HoloLens via a UDP connection over a Wi-Fi network. Near-miss events are identified using a location-based analysis executed directly on the HoloLens. Validation through two experimental scenarios demonstrated a detection accuracy of 91.3\% and data transfer speed at an average latency of 180 milliseconds, which is relevant to a real-time response. The multi-user communication module also achieved a latency of approximately 103 milliseconds. Qualitative insights gathered from nine semi-structured interviews with construction practitioners further confirmed the system’s practical effectiveness in enhancing situational awareness, supporting user response, and improving coordinated safety management in dynamic construction environments.