Automated Object Recognition in Veterinary and Human Medicine

Outset

Development of software for the efficient automated evaluation of a large number of images
© Fraunhofer IVI
Development of software for the efficient automated evaluation of a large number of images

Medical imaging plays an increasingly important role in both veterinary and human medicine. In the diagnostic evaluation of the images, a recurring task is to distinguish and isolate relevant objects and/or regions (e.g., a specific area of a patient’s skin or an animal’s claw) from their surroundings. This step, which is also called segmentation, is often the prerequisite for the necessary determination of quantitative values such as an object's mean temperature.

The larger the number of images to be segmented, the greater the need for an automated segmentation solution. However, it is often a substantial challenge to ensure that the automated segmentation process produces results that are sufficiently accurate in comparison to manual segmentation.

Developments and Expertise

In the course of several research and development projects, new solutions for the automated segmentation of veterinary and human medical images have been developed on the basis of approaches from the areas of traffic monitoring and process observation. A focus area in this respect are studies on the infrared-based measuring of temperature properties in animals, especially in dairy cows.

For the temperature monitoring in dairy cows, solutions allowing object recognition under regular operating conditions were developed and subsequently tested extensively under practical conditions. With the help of learning-based recognition processes with good generalization properties such as »Active Shape«, further developments have made it possible to master even difficult imaging situations with moving objects and changing conditions. For simpler scenarios in animal monitoring, an exceptionally fast segmenting process was developed in cooperation with the KIT in Karlsruhe on the basis of the »level set method«.

Additional methods based on »Active Shape« are currently in use within several interdisciplinary research projects for the recognition of specific areas of a person's face.

 

Overview of Competencies

  • Design of the measurement setup (field of view, camera, lighting, detection of interfering factors)
  • Methods of automated image segmentation
  • Consideration of infrared-specific properties
  • Analysis of image sequences containing moving objects
  • Setup and management of image data bases
  • Development of graphical user interfaces (GUI)
Automatic segmentation and tracking of a dairy cow in a rotary milking parlor (infrared image)
© Fraunhofer IVI
Automatic segmentation and tracking of a dairy cow in a rotary milking parlor (infrared image)
Testing of a method for the automatic recognition of the spinal line in dairy cows (infrared image)
© Fraunhofer IVI
Testing of a method for the automatic recognition of the spinal line in dairy cows (infrared image)

Application Examples

  • Developments for VIONA research and development project
  • Spinal line recognition and measurement uncertainty consulting for AgReseach, DAL and InterAg (New Zealand)
  • Automated detection of facial areas within the cooperative CardioVisio project (with IBMT of the TU Dresden and cardiology center of the Universitätsklinikum Dresden)
  • Quantification of extracellular DNA in fluorescence images of wound models (Institut für Festkörperelektronik of the TU Dresden)
Automatic recognition of facial areas for the analysis of cardiovascular processes (near-infrared image)
© Fraunhofer IVI
Automatic recognition of facial areas for the analysis of cardiovascular processes (near-infrared image)
Detection of extracellular DNA in a wound model (in cooperation with the Institut für Festkörperelektronik of the TU Dresden)
© Fraunhofer IVI
Detection of extracellular DNA in a wound model (in cooperation with the Institut für Festkörperelektronik of the TU Dresden)

Range of Services

  • Engineering in the fields of measuring concepts and automatic image processing
  • Monitoring and execution of measuring campaigns
  • Development of specific image processing algorithms
  • Setup and management of data bases for images and results
  • Development of customized solutions