Computer Vision applications for the industry
Industrial applications and market overview of Deep Learning and Computer vision
Overview
This article gives an overview of the growth factors and drivers of computer vision, the market segments and leaders, and finally concludes with use cases that include the latest advancements in the construction industry, manufacturing industry, and healthcare.
Computer Vision (CV) is a sub-field of artificial intelligence (AI) which consists in bringing human vision capability into computing systems. It deals with interpreting real-world scenarios which are captured by camera-enabled mobile devices in the form of images and videos. Some of the most commonly used computer vision-based applications are facial recognition, human-computer interface, gesture recognition, visual quality inspection of goods in manufacturing processes, in navigation for autonomous vehicles, medical image analysis, and image restoration.
Growth factors and drivers
The main drivers which pushing CV-based applications into the market are the following:
- There is a persistent demand in the market for industrial-based (e.g., quality inspection, predictive maintenance, automation, security, human-machine interaction, etc) or consumer-based (e.g., augmented reality (AR), virtual reality (VR)) applications.
- In the last decade, there has been substantial advancement in technology from both aspects of hardware and software. Machine learning-based approaches, especially deep learning, have seen tremendous growth in the last decade. Deep learning-based models (e.g Convolutional Neural Networks and Transformers) have achieved even accuracies in certain specific tasks(e.g., image classification) higher than humans. Similarly, AI-enabled processors are having a much bigger presence in nearly all mobile devices such as smartphones, drones, raspberry pi, and similar small board computers and consumer electronic devices. It is now possible much more easily process and analyze images on these devices.
- Lastly, all the major software and tech giants like Google, NVIDIA, Intel, Microsoft, Amazon, etc. are increasingly investing in research and development projects associated with CV and AI. There is a huge rush of investments into start-ups that are providing cutting-edge solutions. This money either comes in the form of investment by venture capital firms or in the form of grants by government agencies.
Despite big hype and the success of CV/AI, there remain, however, some challenges for the technology adoption in the industry. Some of them are
Lack of industrial digitization.
Not enough technical experts available in the market.
Concerns regarding data privacy and
Unreliability of the AI algorithms due to their black-box behavior, in other words the lack of model explainability.
Market segments and leaders
The Global AI in Computer Vision market is majorly segmented in terms of components, verticals, and geographical regions. On the basis of components, it is further divided into hardware and software. On the basis of verticals, it can be divided into manufacturing, automobile, healthcare, logistics, security and surveillance, robotics, consumer electronics, entertainment, and many others. On the basis of geographical regions, it can be divided majorly into North America, Europe, and Asia. This market is mainly dominated by North America, especially the U.S., followed by Europe (U.K., Germany, France) and witnessing a fast and rapidly growing rate in the Asia Pacific (China, Japan, India). Some of the major important market players are Google, Amazon, Nvidia, Facebook, Microsoft, Uber, Cognex Corporation, Qualcomm, Apple, Basler AG, etc.
Use-cases
Below are some industrial-specific use cases which demonstrate potential uses of computer vision-based technologies. Some of them are research-type while others are already being used in a production setting.
Construction industry
Computer vision can potentially bring a lot of added value to the construction industry by providing real-time monitoring of construction sites. Another major concern in this industry is the worker’s accidents and mishap happening which can be prevented by monitoring the workers. Here are some of the use cases in this industry:
- Automated progress monitoring of a construction site
Traditionally, an expert needs to visit the construction site and needs to estimate the amount of work done. This process is manual, time-consuming, subjective and prone to error. It is important to compare the actual state of the project to the as-planned state so that potential delays can be identified early within the project life-cycle. This manual process can be automated by using imaging technologies, such as camera, LiDAR and 3D range imaging to capture the current state of the site. Using the imaging data, an estimate 3D model/geometry of the site can be generated which can be compared (or registration) with a 3D CAD model of the planned building. Thus, the current state of construction site progression can be calculated.
- Infrastructure Quality Assessment
It is important on a timely basis to check the quality of the roads, bridges and buildings to ensure the quality to prevent any kind of hazardous accidents. Preventive measures can be taken if some significant cracks/holes can be detected early enough. Due to bigger and long sizes of roads, bridges and buildings, it is not feasible to send a person to manually inspect them and their evaluations can be highly subjective. Therefore, AI can play an important role to detect cracks and holes in the images acquired by a camera system mounted on a drone.
- Workers safety monitoring
The construction company is one of the most dangerous sectors in the world since workers accidents are the highest number compared with other industries. Generally, construction workers accidents are caused by an unsafe act and unsafe condition that can occur separately or together, thus safety monitoring is needed to achieve zero accident. One of the possibilities to reduce the risk of resulting injuries is to ensure that all workers wear the correct Personal Protection Compliance (PPE), e.g., helmet, goggles, gloves and vest. Most of the injuries are as a direct result of PPE failure or non-compliance and these could be eliminated through proper monitoring and alerts. Surprisingly, most businesses still manually monitor employees’ PPE compliance and therefore, there is a strong need to automate this monitoring process with the help of AI. A camera system powered with AI software can be installed at the entrance of the construction site which automatically checks PPE compliance of the workers passing through it.
Manufacturing industry
Quality control, in general, is one of the main challenges that are well-tackled by the usage of AI and computer vision-powered technologies. Traditionally, quality control monitoring is being performed under the guidance of experts, however, it is getting more and more common to rely on computer vision-based systems rather than human visions. These monitoring systems usually consist of cameras with some lightning conditions and computer hardware and software systems.
Key advantages of the computer vision applications in this context are:
improved high-quality control.
decrease in labor cost.
high-speed processing capability and
continuous operations.
Followings are some of computer vision-based applications:
- Packaging (or good) inspection
It is crucial for a manufacturing companies to check whether products that are manufactured are of desired quality and forms. For example, for a pharmaceutical companies, it is important to count tablets or capsules and check their forms (if they are either broken or partially formed) before placing them into containers. Computer vision-based systems can determine whether tablets are broken or partially formed. As tablets make their way through the production line, images of them are taken and are processed to check if the tables are the right color, length, width and not broken. In a similar way, other kinds of goods can be inspected.
- Predictive maintenance
Predictive maintenance is the process of using machine learning and IoT devices to monitor data on machinery and components, often using sensors, to collect data points and identify signals or take corrective actions before components break down. Cameras can be attached to the machines or robots to take pictures of them on a regular basis. These images along with meta-data can be subsequently processed which can eventually identify problems before they happen.
- Defect detection
Manufacturers unarguably want components that roll off the production line to be free of defects and issues, but being able to do this at scale can pose problems for more manual efforts. Machine vision is the ideal technology that can help businesses automate a problem like this. A machine vision-based system can identify defects, store images and associated metadata related to it. As items are processed along the production line, any defects that are identified get classified according to their type, they are then assigned a grade to help further identify the severity of the defect.
Healthcare
With the availability of large medical imaging data, increasing computing power, and decreasing the cost of hardware, Deep Learning has made possible a paradigm shift in the field of medical imaging analysis. Two major concerns in medical diagnostics are
early detection of the disease and
a need for a second reader (or assistant) to complement the main doctor which can potentially lead to the right decision-making.
Some of the specific use cases within the medical imaging (i.e., Radiology) domain are mentioned below:
- Detect Signs of Diabetic Retinopathy
Diabetic Retinopathy (DR) is a diabetic complication that affects the eyes. High blood sugar due to diabetes (both type 1 or type 2) damages the tiny blood vessels in the retina, leading to a distorted vision. Eventually, it can cause blindness. DR is a leading cause of blindness, resulting in up to 24,000 cases each year in adults in the US. To avoid any complication, early detection of DR is important, therefore, patients are advised to have regular screening of their eyes. It is during this stage when the treatments to avoid vision loss are really effective. With early detection, DR can be treated with techniques that have been shown to reduce the risk of severe vision loss by >90%. The AI-based solution can, therefore, be used for an initial screening which can refer detected milder cases to the ophthalmologist. Thus, it can reduce the burden on the ophthalmologists who would have to examine much fewer patients.
- Lung cancer detection using CT scan
Lung cancer is the leading cause of cancer deaths worldwide. A low dose CT (computed tomography) is the standard diagnostic test for lung cancer for those at high risk. Studies have found that screening can either reduce the risk of dying or improve life quality since treatments (e.g., removing the tumor) are effective enough during the early stage. In addition to finding definite cancers, the scans can also identify spots that might later become cancer, so that radiologists can sort patients into risk groups and decide whether they need biopsies or more frequent follow-up scans to keep track of the suspect regions. But, the test has pitfalls: it can miss tumors, or mistake benign spots for malignancies and push patients into invasive, risky procedures like lung biopsies or surgery. And, different radiologists might have different subjective opinions about the same scan. AI may help doctors make more accurate readings of CT scans by looking into complex patterns which might be missed out by them.
- Early Detection of Skin Cancer Detection
Early detection of skin cancer can potentially reduce the risk of advanced melanoma by a high proportion. Characteristics of suspicious moles are described by the ‘ABCDE’ rule: Asymmetry, Border irregularity, Color variation,Diameter and Evolving. Recently, AI-based systems have shown promising results in detecting skin cancer. They were trained on a large collection of dermoscopic images that have been labeled as benign or malignant. However, there are still some concerns related to the reliability of such systems. All the training data were nearly always captured in optimal conditions. Therefore, they cannot take real-world unpredictabilities into accounts, e.g. in a real-world scenario, everyone has a different phone, there are different lighting conditions and background are different and all the racial skin types need to be included in the training data.
Conclusion
We have seen some of the real-world applications of Computer Vision in this article. This list, of course, goes further than that. There is no doubt that Computer Vision and Artificial Intelligence are having more presence in our daily life and will continue to in the future. They can help industries to boost their productivity while ensuring a high quality of the goods. And, soon we will see AI helping doctors to diagnose patients better.
