Realtime Computer Vision

  • Scalable
  • Microsoft Azure IoT Edge
  • Integrated model training
  • Easy deployment
  • Multi-framework support
  • Operation on-premises/in Azure Cloud/hybrid

Realtime Computer Vision

RCV (Real-Time Computer Vision) is a branch of Artificial Intelligence and Machine Learning. It "sees" and understands the content of digital images such as photos and videos. In this context, a distinction is made between the three application areas of image classification, object recognition and image segmentation.

Experience RCV Live at the Demonstrator

Optical Quality Control of Vehicles

The demonstrator in the Microsoft Technology Centre (MTC) uses a model to show how RCV can be applied and used in a running production process.

The test steps for each vehicle model are defined in the test plan and processed sequentially. Test steps include recognition of the manufacturer's logo at the rear of the vehicle, verifying of the vehicle model and model name or inspection of the exhaust trims. Production errors are then displayed with a corresponding message in the front end.

 

Interface between Model Knowledge and User

The dashboard provides the visual representation of the detected object as well as the test results of the deep learning model. It does this by means of so-called bounding boxes. In addition the individual test steps and their results are shown, thus providing information about the reliability and number of objects found. Simple statistics are also displayed, such as the highest ranking tests per day or a bar chart with an evaluation of the tested vehicles per hour.

Technical Specifications

  • Recording images in motion
  • Conveyer belts in circuit without interruption
  • Siemens IPC (Industrial PC)
  • GPU: Nvidia Quattro P2000
  • Deep learning framework with TensorFlow (object recognition, based on Resnet)
  • Allows image evaluation with a processing time of about 1.2 Seconds
  • Model deployment via Azure IoT Edge possible

Microsoft Azure and Azure IoT Edge

RCV from Robotron offers the possibility to provide infrastructure components quickly. It can do this because the specific Azure Resource Manager (ARM) templates are defined for certain application scenarios. These templates are used to deploy predefined PaaS services, e.g. cloudways, storage components, machine learning components, virtual machines (VM) or trigger automatisms. The central Azure service of our RCV solution is Microsoft Azure IoT Edge.

Microsoft Azure IoT Edge provides an innovative approach to system and network architecture helping companies to work beyond the limits of traditional cloud-based solutions. The data generated by IoT Edge devices is typically transferred to cloud servers for analysis. The transmission of large amounts of data is a cost factor to be considered and can lead to time delays or latency due to limited bandwidths. Instead of storing and processing data in a cloud or central data centre, cloud workloads are migrated to the edge of the network. Azure IoT Edge provides features such as authentication and communication from local devices. This allows IoT applications to run either offline or without permanently communicating with the cloud.

 

From the Demonstrator to the Production Line

What works in the Microsoft Technology Centre on the small scale also has numerous application possibilities on the large scale with many advantages for your production environment:

  • Real-time determination of standard deviations by comparing hundreds of images from the production line
  • Screening the assembly for completeness and correctness
  • Ensuring high quality requirements and relieving employees of monotonous tasks
  • Reliable detection of moving objects in various applications along the process chain, including logistics
  • Identification of pseudo-errors
Vertrieb Industrie: Kay Fischer
Your contact person:
Kay Fischer
Sales Consultant Industry