The U.S. construction industry suffers from the highest number of fatalities among all industries, i.e., one in five worker deaths in private industry were in construction industry. There are tremendous lost for their family, for the society and for the contractors. That is the data from the most developed country in the world, consider the situation from elsewhere in the world, the reality could be worse.
One of the top injury causing deaths is from the head, therefore safety issues on the head protection on construction site are very critical. Generally, the guides of working require helmet wearing every second in the construction site; however, due to many different reasons the worker may not to wear it. Plus, except safety concern, there is no usefulness of wearing a helmet.
Our CV technology certainly help the helmet detection more easier than before. The past studies on helmet detection requirement certain images are numerous and most of the images covers enough percentage of human wearing helmet. Traditional CV algorithm requires large number of features are extracted from the images, such as the fringe of the helmet and human head using the gray functions to transform the image into deep black & white, and using Gaussian or related methods to find the gradient of the image. However, since deep learning technologies become popular and well-implemented in the past years. The application of deep learning models in CV has been verified to be better by all means than the traditional methods. For example, YOLO (you only look once) has been large used as object detection methods recently since its great accuracy, speed and robustness in multiclass object detection. The current version of YOLO is V3. YOLOv3 is extremely fast and accurate. In mAP measured at .5 IOU YOLOv3 is on par with Focal Loss but about 4x faster. Moreover, you can easily tradeoff between speed and accuracy simply by changing the size of the model, no retraining required! Most recent researches are on the modification of YOLO on different purpose, different task.
There is another application on helmet safety guarantee issue, which is using VR to enhance the capability of current safety helmet, let the workers aware the working scenario immediately when he/she saw. One great domain service provider is DAQRI (https://daqri.com/worksense/), who considers using AR and VR to help on-site workers understand the working environment, communicate thru helmet to collaborate with remoters.
Both CV and VR solutions require strong ability on video analysis and object detection, especially when you consider a niche market, the objects could be rare and sparse, which asks for special model training.
Technology benefits people. Using CV to automatically monitor the CCTV help the local worker comprehend the potential risk, and simultaneously help the managers to control the workers onsite. If some potential events are triggered, the alerts can be sent to the stakeholders. For application of VR/AR, training workers could be less complicated, all the working operations can be recorded and monitored by the backend system.