: The video frames were used to train YOLOv7 (You Only Look Once) and Mask-RCNN models to detect objects and estimate distances accurately in real-time.
The video is part of a study that addresses the high rate of accidents in the construction industry. Unlike traditional sensors that fire an alarm whenever any object is near, DCAS uses a to evaluate risk dynamically based on:
: Scenarios were built in Unity 3D to mimic real-world construction tasks, such as collaborative excavation. 999 Part 1(1).mp4
: The study noted that moving machine parts (like an excavator's arm) can sometimes obstruct the view or cause perspective distortion, leading to slight distance errors.
: By using the known size of objects and camera focal lengths, the system can estimate the distance of a worker or machine within a small margin of error. : The video frames were used to train
: To save time, researchers used the virtual environment to automatically generate bounding boxes around objects, ensuring high precision for the AI training. Key Findings from the Research
: The system significantly decreased the number of "nuisance" alarms compared to static sensors, as it understands when a worker or another machine is approaching safely for collaboration. : The study noted that moving machine parts
: Adjusts risk based on where the camera is mounted on the machine (e.g., blind spots). How the Video Was Created