Computer vision
Our Computer Vision solutions for industrial sector transform business processes for higher productivity, quality and safety
Dedicated Computer Vision applications for Factory & Process Automation, Quality Assurance and Safety & Security
Top Data Science is utilizing the most feasible and high-performing AI technologies and algorithms for each industrial use case. Productized technology stack enables fast prototyping and fluent scaling to production deployments. We support all common cloud technology platforms including Amazon AWS, MS Azure and Google, and deploy also to on-premise and edge environments.
From new data to right action
Top Data Science solution concept and approach covers the information flow from capturing new data to initiating the correct action based on AI inference. Our data engineering know-how and pipelines enable effective utilization of available image and video data. The Computer Vision applications are trained to identify the desired equipment or process state, which would require a corresponding action to be initiated. Wide range of available APIs and interface services enable advanced integration towards operative systems and processes.
Factory & Process Automation
Quality Assurance
Safety & Security
Even though traditional Computer Vision (CV) has been around, AI-driven CV has brought remarkable efficiency and scalability to industrial automation by replacing rule-based logic with data-driven learning. When coupled with significant advancements in machine learning algorithms and hardware technologies, training, deploying, and serving AI models in production environments have become the lucrative option for automating various industrial processes.
Solutions such as Image Recognition, Object Detection, Semantic/Instance Segmentation, Video Classification, Image Super-Resolution, and Optical Character Recognition provide tools for automating repetitive and mundane tasks.
CV in industry segment and problem focused packages
Even though traditional Computer Vision (CV) has been around, AI-driven CV has brought remarkable efficiency and scalability to industrial automation by replacing rule-based logic with data-driven learning. When coupled with significant advancements in machine learning algorithms and hardware technologies, training, deploying, and serving AI models in production environments have become the lucrative option for automating various industrial processes. Solutions such as Image Recognition, Object Detection, Semantic/Instance Segmentation, Video Classification, Image Super-Resolution, and Optical Character Recognition provide tools for automating repetitive and mundane tasks.
Benefits
AI solutions that enable productivity leap in industrial business processes
Cost savings through higher level of automation
Reduced business and liability risks through improved quality
Improved safety & security of business operations through AI
Testimonial
”Top data science has demonstrated very good AI competence and also very flexible attitude in our cooperation. They have the right attitude to working together with customers for problem solving with AI. Also their team is very easy to work with. I’d warmly recommend them.”
Rashmi Kasat
VP, Head of Digital Business Development, Digital Garage, Metso:Outotec
Learn more about Industrial Computer Vision
What is Computer Vision and why does it bring such great value?
Computer vision is an AI and machine learning technology field that deals with how high-level understanding can be gained from digital images or videos. Traditional computer vision algorithms have been utilized for several years in various tasks in the industry. With recent advancements in deep learning based computer vision algorithms and significant hardware improvements, employing computer vision solutions in production is more lucrative than ever.
AI-based computer vision applications have great potential of improving work quality and experience from many perspectives. They can partially automate some mundane tasks, increase effectiveness, bring consistency to the decision-making, and allow employees to focus on their primary functions. There are needs for companies to develop such applications, share them across their units, easily deploy and operate to bring real benefits to their businesses.
Computer vision applications create competitive advantages for companies in a broad range of industries: from healthcare to public safety; and from retail to manufacturing. In the context of industrial process operation, operators use computer vision applications to retrieve relevant information from the visual data and promptly react to changes in the environment.
What is a platform approach in this computer vision context?
This question is a very understandable one, the world is quite full of platforms. Top Data Science has been helping companies, with our best practices, to accelerate the utilization of computer vision applications using a platform approach. We build and deliver such platforms that can run on the top of AWS, Microsoft Azure or Google Cloud Platform and utilize the full capabilities of such ecosystems. A key benefit we provide here is that the approach is cloud-platform agnostic so the organization can utilize the best of all above mentioned technology platforms. Another key benefit is that we make the development, management and deployment of computer vision applications very easy and cost-efficient for the customer organization.
A platform with easy-to-use graphical user interface can enable industrial domain engineers, without coding efforts, to implement multiple computer vision / machine learning model development tasks in a faster manner.
A platform firstly provides a marketplace of applications and solutions to share them easily within a company’s organization (Good graphical description of this in our whitepaper: Accelerating Computer Vision Utilisation in your Organization). This will help companies be aware of all the existing computer vision activities, and avoid redundant work and investment. Computer vision / machine learning models that are trained by the platform or imported from external environments are visible to users with all needed information such as description or technical details. All the units within an organization can browse through the organization’s solution marketplace to be updated with latest developments and easily find their computer vision / machine learning models of interest, as well as to test models with their data to understand how the models work. If they have specific needs they can fine-tune the models with their own data.
When users start a computer vision / machine learning model deployment process, the models, inference scripts and all the running environments are automatically packaged and ready to be downloaded to run as dockers in local environments. Such implementations are simple and convenient but powerful enough. And all the resources are only used when needed to minimize the implementation cost.
Could you be more specific with the value provided?
We have learned through customer cases that these are some of the materialized value points from computer vision and machine learning based automation:
- Cost Savings: Automating quality assurance processes results in significant reduction in costs by freeing up manual inspection resources.
- Reduced Faults: While quality assurance reduces the prevalence of faulty products or wrong assemblies, with computer vision based solutions the performance of quality assurance itself increases significantly as well.
- Scalability: Unlike traditional computer vision systems, deep learning based computer vision solutions do not rely on hand-tuned rule-based logic. As visual attributes (contours, shapes, textures, etc.) for quality assessment are learned from raw images in a data-driven manner (e.g. supervised learning), our solutions have high scalability and generalizability.
- Standardization: With the help of computer vision and machine learning, quality monitoring becomes objective, reproducible, and traceable. Person-dependent, subjective quality assessments are avoided and consistency is established.
How do you customize and scale computer vision solutions for a specific use case?
Our solutions, both computer vision algorithms and production software, are developed with flexibility and scalability in mind. In our computer vision algorithms we utilize deep learning to avoid traditional rule-based logic. This ensures high generalization power across different welding types, modular design, and less technical debt.
Our production software enables traceability and intuitive monitoring of the quality assurance process as well as possibility of extracting detailed reports. Whole solution can be deployed to any infrastructure, local or cloud (including platforms such as Amazon Web Services, Microsoft Azure, and Google Cloud Platform).
Could you provide examples where computer vision is improving safety?
Securing safety of business operations is of highest importance for any industrial company. Many times it requires a lot of attention, different tools and efforts to cope with. Safety of people moving outside, within e.g. in infrastructures and sites, has to be secured in any condition.
Many of the currently used solutions result in making false alarms due to difficult weather and lightning conditions. This makes the work of control room personnel very challenging. They need to make repetitive site visits to verify the reason for an alert, or that it was a case of a false alarm.
Modern computer vision and machine learning solutions that we have delivered, that are utilizing deep learning, have several advanced capabilities to solve these challenges. These include:
- Ability to identify a human from distant
- Ability to manage and identify distractions like animals, shadows and light reflections
- Ability to take into account difficult weather conditions like snow, rain, night, mist, haze etc.
- The computer vision and machine learning solutions use live stream video data, which enables instant risk detection
- The solutions preserve privacy by enabling detecting a human, but not identifying the person
The benefits achieved through computer vision and machine learning solution deployments have included:
- The solution enabled carrying out the surveillance work remotely, which enabled cost savings and improved the personnel work conditions and well-being
- The solution increased human safety of the operative personnel, not having go and do the site checks, as well as of persons, who might enter these dangerous areas
- Both of these benefits are especially relevant as the deep learning based computer vision solution works reliably in difficult weather conditions
How do you start with computer vision? What does it take to scale the utilization?
Companies identify business use cases where they plan to apply computer vision. Already at this point it is beneficial to have initial discussions with AI experts that will help in identifying such business problems and use cases that match with computer vision type of automation approach. There exists a plethora of applications already in the market so evaluating what approach and tools would work best can be fast and straightforward.
When the business problems and use cases have been selected a logical next step is to talk about the data e.g. what data is available, in what format and environment etc. It makes sense to make a quick assessment with an AI expert on the current data status to validate the discussed AI / machine learning approach. Still there is no definite restriction here not to start utilizing computer vision technologies. There are a wide range of data sets available in the market and collecting image and video data can be many times easily established. Also the data annotation tools are easy-to-use and interoperable with different cloud platforms.
Companies start their computer vision utilization journey many times with quick proof of concepts to evaluate and verify the potential business outcomes. They develop promising ideas further to production-level applications, deploy, operate, and roll-out across units within the companies. This journey we at Top Data Science call co-creation as it nearly always contains combining customer’s domain understanding with our deep AI, machine learning and computer vision expertise. Walking or running through this development path step-by-step secures great results. That is how companies can start utilizing the AI based computer vision applications in their daily work.
Why is computer vision a good tool for automating Quality Assurance?
AI-based computer vision solutions for quality assurance have been proven to create concrete and measurable business value in today’s world. Such solutions can either partially or fully automate various time-consuming and error-prone manual inspection tasks. Scalable nature of these technologies enable significant cost savings, productivity improvements, and standardizing quality monitoring processes.
Companies that need to monitor the quality of their product lines or assembly processes are moving more and more away from manual monitoring that relies on human visual inspection. Top Data Science helps our clients in this transformative journey with accurate and scalable quality assurance solutions. Our industrial computer vision / machine learning technologies can be easily deployed on top of AWS, Microsoft Azure, or Google Cloud Platform and can utilize the full capabilities of such ecosystems.
Example of using computer vision and deep learning for automating Quality Assurance?
This is a very good approach to get a grip of how this is done, so let’s walk through how we have provided computer vision and machine learning based automation for welding Inspection, which requires very high-level accuracy and reliability from the solution.
Welding is a significant part of various industrial applications and especially the automotive sector has several of them. For instance, various components in cars such as rails under car seats have welded metal parts. As these welded parts are exposed to different forces during their lifetime (e.g. shear, moment, vibration), weld seams need to satisfy certain physical measurements for ensuring safety and quality. These measurements are specific to each weld type and there can be hundreds of different weld types in several production sites. This quality assurance process is very time-consuming as it requires several manual measurement steps on digital images. Furthermore, there is significant measurement variability between different human operators which leads to lack of standardization and coherence.
We are proud to have Brose, leading mechatronic component producer and supplier in the global automotive industry, as our customer who currently utilizes our solutions.
We utilize advanced deep learning based computer vision algorithms, i.e. deep neural networks, to perform instance segmentation of various objects (weld seams, welded parts etc.) in captured weld seam images. Our computer vision models are trained in a supervised manner and are customized for high fidelity contour details, batch processing, and fast inference. We have deployed and operationalized our computer vision models in the cloud and also developed user-friendly software solutions for operators and line managers. Our solution grants flexibility and control to our clients when monitoring the welding quality in various production lines in several countries.
Can computer vision be used for automating quality assurance for electronic assembly?
Electronic products are composed of various small components that need to be assembled in a correct manner and in correct order for proper functionality. Typical quality assurance of the assembly process includes manual inspection of the presence of various electronic components as well as other mechanical parts such as screws and bolts. This process is error-prone and very time-consuming. We provide effective computer vision solutions to enable high quality quality monitoring for our customers.
By employing deep learning based computer vision and machine learning e.g. object detection algorithms, we enable highly accurate monitoring of electronic assembly processes. Our computer vision solutions focus on two key features: speed and accuracy. In typical assembly environments visual inspection (manual or automatic) of the product parts has to be performed in a fast manner. Running cutting-edge computer vision and machine learning models with low latency in real-time without sacrificing detection accuracy is an engineering problem we successfully tackle. Furthermore, our solution enables fast training of computer vision models with realistic synthetic images to increase the robustness of the solution in varying lighting, camera angle, camera distance, and focus level conditions.
By developing cutting-edge deep learning based computer vision technologies, we provide accurate monitoring of the electronic assembly process for our clients. This leads to less errors and faults in the assembled products which translates to higher product quality, higher customer satisfaction, lower manual inspection resources and lower repairing and replacement costs.
How do you develop, maintain and operate computer vision solutions in production?
While starting the development of a computer vision solution certain key assumptions have to be made. In the case of machine learning used, e.g. it has to be scoped and decided what data will be used for training and what for evaluating the solution.
The ML Ops life cycle is describing pretty well what all things need to be considered when an AI / machine learning / computer vision solution is taken to production in a way that enables it to be reliable and high-performing.
The life cycle includes the following components (terms may vary per description):
- Data and model management. In the center there are the approaches, tools and procedures to manage both data and computer vision models.
- ML development. The development environment to develop needed detection, classification, segmentation and other computer vision and machine learning models for specific use cases and operative needs.
- Training operationalization & continuous training. In machine learning and computer vision it is essential to build a high-performing, flexible and productized training environment. It needs to support both new model training and continuous training when going to production.
- Model deployment. When a computer vision application and machine learning model has been developed and it is taken into production use, the model deployment comes into play. It has to cater the deployment scenarios for each operative case. Covering both cloud and edge deployments is a more and more common requirement in industrial context.
- Prediction serving. The computer vision application and machine learning model is designed to perform a certain task being e.g. detecting, classifying and segmenting based on new data. These predictions are served to a process, application or solution.
- Continuous monitoring. When a computer vision application and machine learning model is validated to perform against the set criteria being accuracy metrics or alike, it can be taken into production use. Nevertheless a good practice is to continuously monitor its performance to secure that the criteria is met in changing operative conditions. This step prevents concept drift and performance degradation.