There’s also been a move towards digitizing the industry and introducing artificial intelligence across the entire value chain.
The use of AI in the construction industry spans across the entire lifecycle of a project — from forecasting prices of raw material before building starts to predictive maintenance after a building is complete.
Companies like Built Robotics and Construction Robotics produce upgraded heavy machinery with computer vision and AI guidance systems. Their autonomous machines are able to take over the most back-breaking portion of building, thus enabling faster and safer construction processes.
Other companies like Doxel combines AI and computer vision to provide real-time construction site analytics. Their technology helps monitor progress, quality, and safety, providing insights to improve project management.
Two things that affect margins in the construction industry are rework and warranty claims. Some industry insight reports have indicated that rework can account for up to 20% of the cost of a project and that the same issues leading to rework also lead to warranty claims.
Factors that contribute to the need for rework include misinformation during engineering, inefficient material supply, human resources and other communication issues. These are all things that AI can assist with.
AI can help with predicting raw material requirements and placing orders. It can also assist with scheduling and quality control. It can assist with the more mundane or risky parts of construction.
Robots can assist its human operator by automating the more routine tasks, allowing the operator to direct his attention to more complicated and higher value parts of the work. It can also assist with safety and progress monitoring by scanning the construction site and comparing it against the building plan.
Two methods of data labeling are typically used to produce the data required to train autonomous machines and monitoring tools — polygon or bounding box annotation and image segmentation.
In polygon or bounding box annotation, data annotators look at pictures and draw a box around the objects that they want to annotate within a specific picture. After drawing the box, they will have to select from a list of labels to provide an attribution for the object within the box.
These boxes can be 2D or 3D, where the 3D version would also provide information on the approximate depth of the objects. In order to improve accuracy, data annotators will have to ensure that the anchors of these bounding boxes line up as closely as possible with the edges of the objects within the images.
The data produced through this method is also used for safety monitoring of site workers, as well as for a machine to recognise obstacles and thus, be able to navigate within a construction site safely while getting work done in an efficient manner.
For further accuracy, especially in use cases like performance monitoring, image segmentation can be used for feature extraction to add another level of detail to the data.
While bounding boxes provide a set of coordinates that indicate that there is an object within its lines, image segmentation “creates a pixel-wide mask” for each of the objects. Training machines with this data would allow them to pick up more specific information on objects within the construction site.
When it comes to data labeling, having the right tools for the process is vital. Construction may be a single industry, but the variety of AI solutions within just this one industry are vast. There could be a range of different use cases, specific to each individual client.
SUPA’s annotation tool is highly customizable to fit your use cases. Book a demo today.
Everything from uploading data to seeing it labeled in real time was really cool. This is just way simpler to use compared to Amazon Sagemaker and LabelBox. I was also very impressed with how the platform delivered exactly what we needed in terms of label quality.
I was also able to view the labels as they were being generated, which gave me quick feedback about the label quality, rather than waiting for the whole batch. This replaced my standard manual QA process using external tools like Voxel's Fiftyone, as the labels were clear and easy to parse through in real-time.