Evgeniy Krasnokutsky, AI/ML Team Leader, PhD at mobbed
Computer vision is one of the most popular AI technologies whose task is to understand and interpret information from digital images, videos, and other visual inputs.
These properties open up a lot of business opportunities and in this article, we will talk about the most promising applications of this technology.
Medical Image Processing in Healthcare
Sophisticated medical imaging is the driving force behind the power of modern medicine. Traditionally, human professionals perform analysis of medical imagery such as X-rays and CT scans.
The introduction of AI medical image analysis assists physicians in rapid and accurate diagnostics, making inferences based on multiple data points that might otherwise be missed by a human agent.
For instance, Image analysis for cancer detection generally depends upon the skill of the human professional involved in the assessment, but computer vision and machine learning implementations make it easier than ever for professionals to identify difficult-to-find signs of cancer.
In addition to the identification of cancers, AI for medical image analysis also assists human agents in detecting diabetes, helping with complex surgeries, providing additional support in oncology, medical laboratory work, and even predictive diagnosis to help cut down the time of and improve long-term care for patients.
According to the Oxford Academic in 2020, machine learning assists technicians with the difficult task of bacteria counting in combination with plate methods.
Convolutional neural networks provide superior image analysis capabilities, leveraging their unique advantages with larger datasets to more efficiently extract quantitative properties from medical imagery.
Because of the power of AI integration in medical image processing, computer vision and machine learning implementations are vigorously employed both in the field and in comprehensive medical research.
Defect Detection in Manufacturing
Quantifying the actual costs of product defects is difficult, especially when considering the direct costs of such products, including notifying customers of the defect, repackaging and transporting products, destroying or disposing of any defective products, and replacing customers with defect-free products. According to Allianz in a 2020 survey, product recall events cost companies almost $50 billion in expenses.
Computer vision is a serious investment for manufacturers looking to reduce defects in products, mostly because it is far superior to the error-checking capabilities of human agents. Car insurance firms use computer vision to assess damage in claim settlements.
Using a simple Android phone camera, one such company was able to estimate car damage from multiple angles with an inference time of just two seconds, and achieved an astounding accuracy rate of 97 percent.
Machine learning-based visual quality inspection of products is more efficient than human agents as inspecting large-scale production lines and determining even the most difficult-to-find faults in the final product.
Machine learning defect detection has a wide range of applications from identifying natural defects in raw materials to identifying potential break points in automotive and electronic parts.
Self-checkouts in Retail
When Amazon acquired the popular organic and health foods retailer Whole Foods, the company made headlines by introducing its Amazon Go checkout scheme.
Amazon Go utilizes computer vision algorithms as the core of its “Just walk out” tech to give customers a convenient shopping experience, all while improving loss prevention schemes and collecting valuable metrics for marketing purposes.
Rather than using a self-checkout or depending on human clerks, the Amazon Go system provides a seamless shopping experience, identifying which products were purchased by the customer based on live video feeds, product weights, and positional data.
Theft is one of the leading causes of loss to retailers, with over 2 percent of all thefts occurring in retail, with an unknown number of thefts occurring due to self-checkout.
Self-checkout was initially introduced by retailers to provide shoppers a more convenient shopping experience while also cutting down on the number of human agents needed for checkout positions, but it has also caused untold losses in profits due to the ease of theft.
Adopting machine learning helps retailers identify which products are most at risk of theft, and also assists security officers with real-time assessments to apprehend in-field theft.
Automated License Plate Recognition (ALPR) in Automotive
Manually identifying license plates proves a difficult task for human agents, especially since it depends on natural limitations such as the reliability of the agent’s natural eyesight, prone to human error. Automated License Plate Recognition or ALPR is utilized by law enforcement agents to identify license plates automatically.
ALPR uses computer vision-enabled vehicle-mounted cameras as well as mobile device cameras to intelligently detect license plates. The technology can also be used in combination with security cameras or with still images to identify vehicle plates.
Vehicle-mounted ALPR systems leverage useful data such as current GPS coordinates, information about the color and model of the vehicles, contextual information such as the vehicle’s surroundings, and more. Officers can use this data to identify potential drug dealers and human traffickers by their traffic patterns.
Parking enforcement companies such as LAZ Parking have also hopped on board with ALPR technology to identify parking occupancy, utilizing vehicle-mounted ALPR devices to check whether vehicles have paid up for parking.
These parking companies have taken inspiration from models like Amazon Go, moving to subscription-based parking models where customers opt-in for parking by downloading an app to seamlessly pay for parking before driving into the lot.
This is Not the Limit
The sky is the limit when it comes to business applications for computer vision systems. Some of the most interesting implementations of computer vision include crop growth monitoring and plant disease identification, which utilizes image feeds to make predictions as to whether or not particular crops are susceptible to diseases and changes in the watershed.
Plugging into larger systems such as geological and meteorological data, these systems can leverage an incredible amount of data to do predictive work that exceeds the capabilities of hundreds of highly-trained human analysts.
The modular nature of these systems is what truly takes them to the next level, and AI is sure to revolutionize every field without exception in the coming decades.