Master the distinction between computer vision and machine vision. Discover which solution, custom or COTS, fits your machine vision project needs best.

What is the difference between machine vision and computer vision?

Machine vision (MV) and computer vision (CV) are both fields that deal with processing visual information, but they have distinct applications.

Computer Vision is a field of artificial intelligence that automates the capture and processing of images, with an emphasis on image analysis. Its goal is not only to see but also to process and provide useful results based on the observation. Computer vision can be used to analyze various visual data, including images, videos, and even 3D point clouds, to extract meaningful information.

Machine vision, a subfield of computer vision, is specifically designed for applications in industrial settings. It often leverages computer vision techniques for tasks like object detection, classification, and tracking.

The key difference between these two fields lies in their application. Computer vision has a broader scope and finds applications in diverse areas like digital marketing and autonomous vehicles. In contrast, machine vision focuses on specific industrial tasks such as quality control and automation.

EnCata Product Development, a hardware product development service company, specializes in machine vision solutions.

Structure of the machine vision hardware

Resource

Machine vision hardware relies on several essential components:

  • Lighting: Illuminates the object being inspected to highlight its features for clear camera capture.
  • Lens: Captures the image and focuses the light onto the sensor.
  • Image Sensor: Converts the captured light into a digital representation, similar to our eyes converting light into signals for the brain.
  • Vision Processing (VPU): This brain of the system employs algorithms to analyze the image, extract relevant information, perform inspections, and make decisions based on the extracted data.
  • Communications: Enables the system to transmit information either through discrete I/O signals (on/off switches) or by sending data over a serial connection to a device that logs or utilizes the information.

Furthermore, most components, like lighting modules, sensors, and VPUs, are readily available as Commercial Off-the-Shelf (COTS) options. This allows for building custom systems or purchasing pre-configured ones with all components integrated.

Custom vs commercial off-the-shelf (COTS) solution

Choosing between building your own hardware (custom) or buying a pre-made option (COTS) for your machine vision needs depends on several factors:

Time to Deployment

  • ‍COTS: Offers the fastest path due to immediate availability. However, you will need to ensure compatibility with existing systems.
  • ‍Custom: Requires significantly more time due to various development stages (planning, design, testing, etc.), often exceeding estimated timelines.
  • ‍Low-code/no-code: Provides the quickest solution, allowing app creation in potentially a single day. AI advancements like Microsoft's Power Platform further accelerate development.

Flexibility

  • ‍COTS: Offers a one-size-fits-all approach, potentially limiting customization. Modifying them can be expensive.
  • ‍Custom: Provides ultimate control over design, technology, and features, but may suffer from communication gaps between business and technical teams, leading to technical debt.
  • ‍Low-code/no-code: Balances flexibility with technical debt management. Citizen developers within your organization can create intuitive apps with less reliance on IT teams.

Affordability

  • ‍COTS: Available on subscription or perpetual license models, providing flexibility in payment.
  • ‍Custom: Can be more costly due to the extensive development process.
  • ‍Low-code/no-code: Offers an affordable alternative, especially for quick solutions that don’t compromise flexibility.

Ultimately, the choice between COTS, custom, and low-code/no-code development hinges on three key factors. By carefully considering them, you can make the best decision aligned with your organization's unique needs.

Custom machine vision solutions excel in various scenarios where commercial off-the-shelf (COTS) options might fall short:

Highly Specialized Tasks

  • Tailored for unique needs: Custom solutions excel in handling tasks with specific requirements. For example, in manufacturing quality control, custom systems can be fine-tuned to detect defects unique to your product, unlike generic COTS solutions.
  • Optimizing agricultural processes: Custom vision systems can be designed to identify specific crop diseases or pests, leading to improved agricultural yield.
  • Enhanced medical diagnostics: Custom algorithms can be developed to analyze medical images with greater accuracy, focusing on specific conditions or rare anomalies.

Complex Environments

  • Resilience in harsh conditions: Custom hardware can be built to withstand challenging environments where COTS solutions may fail. This includes extreme temperatures encountered in industrial settings or outdoor applications.
  • Space exploration: NASA's Mars rovers rely on custom vision systems to navigate the Martian terrain, adapt to dust storms, and analyze rock formations.

Performance Optimization

  • Real-time processing: Custom solutions can be optimized for speed, enabling real-time processing crucial for applications like autonomous vehicles.
  • Efficient resource utilization: Custom vision systems can be designed to operate efficiently with limited computational resources.

Privacy and Security

  • Enhanced data control: Custom solutions offer greater control over data privacy and security. This is particularly important when dealing with sensitive data, such as facial recognition in banking applications, where custom solutions ensure data remains within your infrastructure.
  • Closed-loop security: Custom hardware can operate in air-gapped environments, minimizing the risk of external security breaches.

Integration with Existing Systems

  • Legacy Systems: If you have legacy machinery or software, custom solutions can be seamlessly integrated with your existing infrastructure.
  • Industry-Specific Requirements: Custom solutions allow to meet the industry standards and protocols.

While custom solutions offer significant advantages, they also come with higher development costs, longer lead times, greater need for specialized expertise. Carefully consider these factors alongside your project's unique needs to make an informed decision about whether a custom solution is the best fit.

What solutions are suitable for machine vision hardware prototyping?

  1. NVIDIA Jetson Nano: This powerful and compact option excels in AI-powered computer vision tasks like image classification, object detection, and more. It supports open-source software and libraries like OpenCV and benefits from a large developer community. Additionally, it finds use in real-world applications with readily available projects.
  2. Raspberry Pi: This versatile and affordable series of single-board computers enjoys widespread use in various fields, including machine vision, due to its strong community support and wide range of capabilities.
  3. PINE A64-LTS: This ultra-low-power and cost-effective option offers 64-bit performance, making it suitable for various applications, including machine vision.
  4. ASUS Tinker Board: This compact board delivers robust performance for its size and is capable of handling machine vision tasks.
  5. LattePanda: This small single-board computer with a built-in Arduino-compatible co-processor can be used for machine vision applications.
  6. ODROID: ODROID is a series of powerful single-board computers and tablet computers created by Hardkernel Co., Ltd. It’s capable of running multiple operating systems including Android, Ubuntu, and other Linux distributions. 
  7. UDOO: These open-hardware boards offer a flexible environment for exploring machine vision concepts.
  8. Arduino: While less powerful than other options, Arduino boards are popular in education and prototyping and can be used for simpler machine vision tasks.

Note that the specific choice of hardware can depend on the requirements of your machine vision task, such as the complexity of the vision algorithm, the required processing speed, power consumption, cost, and other factors.

What defines the necessary capacity of machine vision hardware

The necessary capacity of machine vision hardware is determined by a combination of hardware specifications, system performance, and application requirements. There are main hardware factors:

  1. Sensor Resolution: Think of the sensor as the "eye" of your system. It captures light and converts it into a digital image, similar to how our eyes work. Sensor size determines the area the camera can "see" at a given distance. Pixel resolution determines the level of detail the camera can capture. The higher the resolution (more pixels), the sharper the image and the smaller the details it can detect. As an illustration, for industrial inspections requiring high precision, a camera with at least 5 megapixels might be needed to identify small defects on manufactured parts.
  2. Lighting: Unlike human vision, machine vision relies on analyzing reflected light to understand its surroundings. Therefore, lighting is crucial for capturing clear and accurate images. Proper lighting setup can enhance specific features of interest, minimize unwanted details, ensure consistent illumination across the entire image. Example: a quality control system for printed labels might require bright, even lighting from LED lights with an intensity of at least 1000 lux. This ensures accurate detection of defects.
  3. Performance Losses: Machine vision systems rely on algorithms to analyze captured images and make decisions. However, inefficiencies in the algorithms or hardware can slow down processing, impacting the system's performance. In a robotic application, the hardware should have a processing time of less than 50 milliseconds per frame for smooth object identification and robot movement.
  4. Downtime: Machine vision systems can also experience reduced capacity due to downtime. This downtime can be caused by several factors, such as mechanical failures in the hardware, software bugs that disrupt operation, or even human errors during operation. For instance, a dirty or misaligned camera lens can prevent image capture, hindering object identification and ultimately impacting downstream processes like robot guidance.

In closing,

Machine vision hardware consists of components like lighting, lenses, image sensors, vision processing, and communication. These can be either commercial off-the-shelf (COTS) or custom-built. Custom solutions offer flexibility, scalability, and alignment with unique business needs. They adapt to evolving requirements, unlike one-size-fits-all COTS products. Customization ensures optimal performance and long-term viability. Machine vision hardware is a critical component of any machine vision system. It determines the system’s performance, reliability and cost-effectiveness. The right choice of components can help you achieve your goals while minimizing risks and costs.