Manufacturing & Digital Workplace

A key use of AI on assembly lines is for the identification and counting of components and the detection of faulty components. However, with the high levels of quality control present in assembly environments it’s often very unusual to see defective products. Mindtech’s Chameleon Platform enables virtual scenes to be created that can be used to generate large quantities of images for object counting and also for the identification of defective products to train an AI network against. Detection of correct health and safety protocols and worker safety

Product Inspection – Rusty Bolts

Use Cases

  • Product Counting
  • Quality control
  • Faulty Item detection

Chameleon Benefits

  • Quality control processes mean faulty products are not often seen on assembly lines. In Chameleon faulty items can be created and procedurally adjusted per run to ensure a large variety of images can be created of faulty products
  • Where real time counting is required clips can be created faster than real time to ensure robust performance of the AI network
  • Lighting, shadowing and backgrounds can be easily adjusted to ensure generalization of the detection to multiple environments

Rusty Bolt Datapack Example

AI Network function = Defective Bolt Detection

Parameter Number Total Clips Comment
Environment 1 1 Conveyor belt
Asset types to identify 10 10 Different types of bolts
Deformations 5 50 5 Bend types
Cracking 5 250 5 crack types
Patination 20 5,000 Rust variation  types
Total Clips 5,000
Frames per Second 2 10,000 Platform typically runs at 30fps
Clip Length (seconds) 10 100,000
Total Images 100,000

Site Health and Safety

Use Cases

Significant health and safety rules govern building sites. Monitoring employees and machinery can be a significant task

  • Hard hat & protective clothing use
  • Safety Perimeter detection
  • Employee counting & Tracking

Chameleon Benefits

  • Building sites tend to be dangerous environments. Dangerous corner cases can be easily created in Chameleon and repeated under differing lighting and weather conditions
  • Unique environmental conditions e.g. dust, can also be created with no risk to equipment or people
  • Delays in construction schedules to create training data can be avoided

Object Detection & Materials Management

Use Cases

  • Lost objects
  • Material checking to avoid theft
  • Safety checking for dangerous materials
  • Material counting and tracking

Chameleon Benefits

  • Misplaced items are a regular issue on building sites. Any item can be imported into chameleon for identification/counting.
  • Where the item is unique digital twin technology can be used to import an exact image of an object into Chameleon e.g. the valve in the images shown
  • Stock counting and checking use cases can be easily created in Chameleon with Lighting, time of dat and weather adjusted to provide a broad range of training data

Machine Safe Distancing

Use Cases

  • Heavy equipment is often deployed on construction sites e.g. cranes, bulldozers. Ensuring that the driver is aware of the proximity of humans to their machines can prevent accidents
  • Out of bound areas can be mapped using AI and site management can be alerted if employees stray into dangerous areas

Chameleon Benefits

  • Cameras can be placed at multiple positions on the heavy equipment to determine the best perimeter view (including fisheye cameras)
  • Range and velocity of the machine to the configurable actor can be provided
  • Dangerous areas can be mapped into a scene and scenarios created of configurable Actors walking through these areas with no danger to employees

Construction Vehicle Examples

Document Digitisation

Use Case

  • GDPR and privacy issues mean using real world documentation for AI network training can be difficult
  • Creating templates of real-world documents populated with synthetic data and images is a powerful way of creating high quality, varied training data

Chameleon Benefits

  • All information is synthetic (face/names/dates etc) so no privacy issues
  • Information procedurally generated per document to ensure randomization and diversity
  • Documents can be lit in multiple ways and with multiple backgrounds to represent how images are provided by the general public
  • Bounding boxes can be customized to text strings or to per character
  • Damage, worn, yellowing documents can be created to broaden the test dataset

Synthetic Document Examples


If you would like to discuss how our Data Packs could work for your organisation, please get in touch.