Whenever you look at something, your brain carries out a series of processes without you even acknowledging it. First,
it controls the eyes in order to acquire the best image then it recognises what is the particular object seen.
Then, it inspects anomalies to reach an OK/NOK decision.
Autonomous Machine Vision – Artificial Intelligence (AMV-AI) follows this same cognitive process. AMV-AI comprises of three
independent and synergetic AI engines driving a unique electro-optic system
Getting the Image
- Self set-up of all optics parameters per specific use case & scenario
- Self adjust mechanism during runtime dynamically mitigating changes in the production environment
- Smartly selects optimized image from the live stream for the recognition and inspection engines
Identifying the Part
- Recognizes and detects rigid objects of any shape and surface
- One reference image is sufficient for detection
- Supports variations in location and rotation in the Field of View (FOV)
- Can be self-triggered or use an external trigger
Finding the Defect
- A novel semi-supervised, non-specific AI technology to overcome specific data scarcity and leverage generalization
- Requires 20-30 good parts (OK) only, with no onsite training
- Understands product tolerances and physical attributes
- Differentiate between defects and permittable defects (without any prior defect references nor training)
- Ongoing improvement via continuous deep- learning
Fully Contained Electro-Optic System
- A multi parameter electro-optic vision system designed for ultimate flexibility
- Embedded smart PWM lighting array with varying light direction and intensities
- Mega pixel global shutter image sensor with optical zoom
- Novel, patent-pending architecture
AMV-AI is not powering a software – it is powering a product
It is not used for the inspection module only, it is an end-to-end AI solution
End to end: from image capturing to inspection
Dynamically adjusts the electro optics system to acquire the best image (from one image)
Automatic independent detection of object
Only need a small amount of OK images. No need for NOK images.
No training process
vision with AI
Image acquisition is pre-set and not dynamic
Requires many OK & NOK objects
Requires training & development of rules for pattern matching
- Embedded illumination is controlled by pulse-width modulation
- Using boosted pulses of illumination synchronized with the camera exposures, taken autonomously at the best moment to capture the object
- LEDs are arranged in several distinct regions, which allow the system to autonomously control the direction of the illumination and take several images with varying light direction and intensities. These are then fused to create a single reflection-less HDR image.
- No flickering or light variation
- Works in real time to make on the fly adjustments based on changes to the exposure, the iris, etc.
The AI engine knows the characteristics of a good part and can deduct a defect part based on this. There is no need for any bad parts during learning.
- No need for not OK (NOK) or defected samples or images
- No need to create defect definition rules
- Accurate identification of the smallest defects
- High sensitivity to identify defects vs. permittable variations
- Industrial grade inspection performance with very few good samples required
- Picks small defects that would be missed by the human eye.
- Identify defects on black surfaces.
- Inspect deep and complex 3D shapes.
- Easily inspect and makes decisions for complex sequence scenarios with multiple inspection points
Our super skilled research team has been working on meta-learning since 2015, when the academic world just started in 2019.
AMV-AI is a novel AI architecture and a unique formulation of neural networks utilising computer vision, deep learning and real-time software optimization technologies.
vision & deep learning expertise