CaRC Lab
Computing and Robotic Construction Lab
Vision-based Construction Site Monitoring
Construction videos contain important visual information related to productivity, activity, safety, and planning. We use various deep learning methods (e.g., object detection, tracking, activity recognition, and instance segmentation) to monitor construction sites automatically from visual data.
Detection of construction objects
Excavator activity recognition
Tracking multiple construction machines
Instance segmentation of construction machines
Progress and Productivity Monitoring
By integrating the detection, tracking, and activity recognition methods, we are able to obtain the productivity of construction workers and the progress of the entire project. This technology has been tested in a real construction sites in Hong Kong with 2 months video.
ChatGPT for Automatic Daily Reports Generation
By bridging ChatGPT and computer vision, we can generate daily reports from construction videos automatically. The generated reports are comparable to human written reports. This technology can save the time and efforts for construction engineers in daily documentation tasks.
Construction Video Highlights Detection
Construction videos are usually very long and the contents change very slowly. By integrating feature extraction and object tracking, we can detect video highlights from long-term construction videos. For example, one hour video can be compressed to one minutes with retaining important project-related information.
Nighttime Construction Monitoring
In the city, road maintenance are usually conducted at nighttime. However, nighttime construction is dangerous due to its limited illumination conditions. Automated monitoring of nighttime construction is beneficial to construction safety. We adopt illumination enhancement method to monitor nighttime construction to recognise machines from videos that are even difficult to be recognised by human eyes.