Institute of

Cognitive Integrated Sensor Systems

Prof. Dr.-Ing. Andreas König

News

Staff

Teaching

Research

Publications

Conferences

Student Theses

Contact

Impressum

Home

CMOS Image Sensors with Integrated Preprocessing for Application-Specific Embedded Vision Systems

Subject:
Investigation of innovative CMOS compatible image sensor principles to achieve high dynamic range and high contrast resolution for machine vision. Integration of efficient analog or mixed-signal preprocessing methods with the image sensor linear or matrix array for compact, low-power, and real-time system implementations.

Abstract:
In this project innovative CMOS image sensor principles are investigated. The objective is to achieve high dynamic range and high contrast resolution as required for real-world problems, e.g., in automotive image processing tasks or industrial image processing for visual inspection applications. Especially for the significant real-time requirements of image sequence analysis problems, the joint implementation of integrated analog or mixed-signal spatio-temporal preprocessing with the CMOS linear or matrix image sensor is investigated. The overall objective is to achieve compact, low-cost, and low-power application-specific vision systems.

Three key applications were identified and selected for the investigations. The first application is overtake monitoring as one example of a driver-assistance component in the automotive segment. The second application is the task of eye-tracking to control a 3D-display and provide a continuous stereoscopic impression to the user. The third application comes from industrial visual inspection and targets on in-line visual inspection, e.g. width measurement or quality control for high speed production processes.

 

In a top-down approach, system simulators for the first two applications were developed. The focus on the selection and developments of algorithms was on their suitability for ensuing VLSI-implementation. Especially the partitioning in an analog or mixed-signal front-end for computationally intensive spatio-temporal preprocessing and a programmable, purely digital dedicated back-end was key constraint in the investigations.

A first vision-chip was implemented in VLSI, using Austria Mikro Systeme Int. AG 0.8 microns CYE technology. It is a linear sensor comprising three rows of 224 HDR-pixels each with global adaptation. The sensor implements adjustable spatio-temporal smoothing DoG computation and edge detection. Further, for visual inspection tasks ROI selection, thresholding, and binarization are available. The vision-chip was packaged in a DIL48 and a prototype camera was assembled.

ffff

A tiny demonstrator with this prototype camera controlled by a PC lab board has been completed and a GUI for chip programming and image acquisition has been implemented. 

To achieve matrix images, as e.g., in visual inspection, a moving object is required for demonstration. We resorted to the application of a model train to obtain images and test the vision-chip functionality. An improved demonstrator was aspired, using a mini module with a TI DSP C50 to allow much faster processing and optional postprocessing on the DSP in a compact module as a next step.

  Status:   concluded, duration 07/1998-06/2001
  Partner:   -
  Financing:   Deutsche Forschungsgemeinschaft DFG, Graduiertenkolleg Sensorik
  Contact:   Prof. Dr.-Ing. Andreas König
  Contributors:   Jan Skribanowitz (Key contributor), Thomas Knobloch, Jörg Schreiter, Jens Döge (student contributors)
  Publications:    
      Skribanowitz, König, A.: Architekturstudie für die mikroelektronische Implementierung eines Überholwarnersystems zur Fahrerunterstützung im Automobil. In Tagungsband Fachtagung Informations- und Mikrosystemtechnik, Magdeburg , S. 249-256, 25.-27. März, 1998.
       
      Skribanowitz, J., Knobloch, T., König, A.: Behavioural Definition of an Application Specific Vision-Chip for Overtake Monitoring and Related Tasks in Automotive Systems. In Proceedings of the EUFIT'98, Aachen, Germany, Sept. 7-10, 1998.
       
      Skribanowitz, J., Knobloch, T., König, A.: VLSI-Architecture of an Application Specific Vision-Chip for Overtake-Monitoring and Real-Time Eye-Tracking. In Proceedings of the 5th International Conference on Soft Computing and Information/Intelligent Systems IIZUKA'98, pp. 72-75, Iizuka, Fukuoka, Japan, October 16-20, 1998.
       
      Skribanowitz, J., Knobloch, T., Schreiter, J., König, A.: VLSI Implementation of an Application-Specific Vision Chip for Overtake Monitoring, Real Time Eye Tracking, and Visual Inspection. In Proc. of the 7th Int. Conf. on Microelectronics for Neural, Fuzzy, and Bio-Inspired Systems MicroNeuro'99, University of Granada, Spain, pp. 45-52, April 7-9, 1999.