There is an abundance of visual data present and some of it even collected, but most of it largely unutilized to its full potential. Are you in a possession of such valuable data? Then continue to read how computer vision can help you.
What is Computer Vision
Computer vision enables automation of the visual understanding tasks from digital images, videos and other visual inputs through the use of artificial intelligence (AI). Many people like to think of AI as being an attempt to mimic human intelligence and computer vision tries to mimic humans ultimate ability to see, observe, understand and act upon. This is not a science fiction anymore.
Massive amounts of available curated imagery data, together with algorithmic advancements in the field of deep learning has enabled computer vision to move from a clumsy manually engineered systems to generalizable data driven systems, which enabled its proliferation and commercialization. Curated largescale imagery datasets with human-labelled objects of interest are crucial as the current commercial computer vision solutions are mainly driven by the supervised AI.
Common ways in which business use Computer Vision
All of the photos that we take with our phones are touched by computer vision, but can we go beyond nice photos?
This is probably the most prominent example of computer vision use in practice and Elon Musk once again promises us that we will achieve self-driving capabilities this year. We cannot promise you that, but if you recently bought a new car, you can most definitely see computer vision at work when using cruise control and decide for yourself. Full self-driving is nevertheless difficult to achieve due to the long tail challenge and the heavy use of supervised AI, which providers still seems to rely heavily.
Manufacturing domain is probably one of the oldest application domains of computer vision, with its roots in quality control (QC), where machine vision and basic image processing have been used for decades to measure product dimensions, shapes of objects or any other basic characteristics that can tell bad produce from a good one. The new paradigm of a data-driven computer vision has been reinventing the domain of QC, by dropping the reliance on complex manually engineered rule-based decision systems, increasing the effectiveness of such systems and their ability to spot even the slightest of product deficiencies, but the current solutions are still heavily reliant on human labels and supervised AI.
Medical imaging is another example of an established domain where computer vision and its basic image processing techniques have been used for decades but are now being revolutionized with advanced deep learning-based solutions to improve the speed and accuracy of the diagnostics, by reducing the burden on scarce availability of health professionals. Carefully developed AI solutions in different application domains have been shown to reach or exceed human level accuracy, while heavily reducing the workload. Despite the success of such solutions, the development of heavily supervised AI solutions in the healthcare domain poses a great challenge, as the complex medical imagery data needs to be labelled by domain experts, which is hard to obtain in large quantities.
CCTV cameras are becoming ubiquitous, not just to capture the bad guys, but also to effectively monitor the infrastructure and to spot events that are important for its safety. What if your infrastructure spans for miles, or hundreds of miles? Did you know that you can monitor your infrastructure from space? The availability of satellite imagery nowadays, combined with computer vision gives an unprecedented ability to detect a change or a particular object of interest and Computer vision will soon enable to Google theEarth through satellite imagery directly.
How we are using Computer Vision technologies?
We are not building self-driving cars at XLAB, but are nevertheles moving the frontiers of computer vision. We are genuine believers that AI should solve real-world problems in a real-world fashion. Our computer vision solutions are striving to reduce the need for labeled data and thus making our computer vision solutions easier to adapt. We are applying solutions for anomaly detection, object detection, segmentation and counting to many different application domains:
Unsupervised Surface Anomaly Detection
Our solutions only rely on defect-free samples, which are trivial to collect on productions lines, where defects occur rarely. We are then able to model the normal appearance and detect and segment any deviations from that normal appearance. This kind of a solution greatly differs from the industry standard supervised solutions and is also able to detect novel anomalies which were not predetermined.
Object Detection and Counting
Our solutions go beyond industry standard bounding box object detection approaches and can accurately detect and classify objects based solely on labelled points instead, which require significantly less effort to obtain. Our award winning solutions have been used successfully to quantify cellular structures in the field of pathology.
Infrastructure monitoring using Earth Observation
Our solutions strive to optimize large-scale infrastructure management by passive monitoring through satellite imagery and the use of computer vision methods for change or anomaly detection, as well as object detection and counting. We are also building specialized solutions for vegetation monitoring in the infrastructure right-of-way.
How can you benefit from Computer Vision?
It’s time to start using your data to your advantage. Our highly skilled AI experts help clients from various industries capture and use data to drive informed strategic decisions, automate processes and detect potential problems before they become major issues. We can effectively apply our expertise and knowhow to any domain and develop a solution that meets your unique business needs. Reach out to make the next big leap with AI.
This work is partially funded by European Commission project iPC (grant agreement number 826121).