Hough transform system enhancements resulting from neural network implementation
G.L. Dempsey, Eugene S. Mcvey · 2002
A neural network implementation of a Hough transform system has been suggested to make real-time operation possible. That and other developments are summarized and combined to suggest a complete system. Convergence times under 30 mu s are obtainable with general-purpose operational amplifiers for the processing nodes or neurons. Enhancements are possible with an artificial neural network system such as multiple target detection with a simple hardware addition to the basic system. Approximation of the peak magnitude in parameter space is available from the multiple peak detection circuitry. The magnitude of the peak indirectly yields the length of the line segments in image space. The system can adapt to varying image intensities. This feature can be used to locate weak targets and avoid neuron amplifier saturation. A signal perturbation technique can be used to enhance resolution or reduce hardware requirements.>