Image processing

Real-time image processing solutions for cost-optimised embedded systems pose additional challenges for developers once a software prototype has been completed and validated: 

  • Optimisation of memory requirements and processor load: Where can image-oriented algorithms be replaced by line-oriented algorithms, up to which processing level is access to the original image data required?
  • Replacing floating point with integer arithmetic: Which scaling and word widths maintain the accuracy of the results?
  • Partitioning of algorithms for parallel processing on multi-core architectures: Where should the system be cut to balance the load between the cores, how much delay does the communication between the cores generate?
  • Proof of compliance with real-time requirements: Can the target system compute its responses at frame rate regardless of the content of the video stream?
  • Proof of functional equivalence: Does the target system deliver the same responses as the prototype for the same video data?

To solve these tasks, close co-operation between process developers, porting engineers and hardware developers is required. FORTecH Software has developed a methodology for the safe and efficient transfer of prototype solutions into production-ready products based on the experience gained from various projects for processing real scenes. Automatic tests guarantee continuous verification of the target code.

Porting prototype solutions

FORTecH Software has developed a methodology for the safe and efficient transfer of prototype solutions for processing real scenes into products ready for series production. After completion and validation of a software prototype, some typical tasks arise during the transition to cost-optimised target systems:

  • Replacing floating point arithmetic with integer arithmetic
  • Partitioning algorithms for parallel processing on DSP multi-core architectures and FPGAs
  • Optimisation of memory requirements and processor load
  • Proof of compliance with real-time requirements
  • Proof of functional equivalence (identical responses with identical input data)

During porting, we work closely with the customer's process and hardware developers. We guarantee functional equivalence between the prototype and the target system through the automated comparison of interim results.

Innovative 3D applications

FORTecH Software has gained experience in several projects with the extraction of 3D information from real scenes using stereoscopic or moving monocular camera systems and time-of-flight sensors. The use of three-dimensional information often leads to surprisingly simple solutions compared to two-dimensional methods. 

Camera calibration

A typical problem in image processing is determining the intrinsic and extrinsic calibration of the cameras used. We have experience in implementing application-specific solutions for both forms of calibration in embedded and cost-optimised applications.

 

Intrinsic Calibration Target

 

Optical tracking

Image processing systems for processing real scenes typically contain a world model. It manages statically validated hypotheses derived from observations of the environment. This approach allows the robust identification of objects from one or more video streams even if the analysis of the individual images only provides uncertain information. FORTecH Software offers a framework that supports the efficient implementation of customised optical tracking solutions. The application-specific system components are integrated into the framework in the form of plug-ins.

Contact