Intelligent Image Processor based on neuron-MOS image sensor merged architecture

点击翻译按钮获取中文摘要

An image captured on an image sensor chip has an enormous amount of information. It is spatially two-dimensional and, if motion is concerned, it becomes three dimensional because time axis is added. It is almost impossible to perform a real-time recognition of the image by a sequential computation on general purpose computers even if it is the most advanced one of today. The purpose of this research is to develop a basic hardware of integrated-circuit systems that can perform real-time intelligent image recognition. Two integrated circuits have been developed that can real-time detect and trace the movement of an object using a high-functionality device called Neuron MOS transistor (vMOS) as a key circuit element. The processing is conducted on two sets of one-dimensional data obtained by projecting a two-dimensional image data onto x- and y-axs. As a result, it has become possible to detect the motion vector between two successive frames within a few hundred nono secondS,which usually takes about a few tens miliseconds using plural of dedicated DSP chips. We also developed a circuit which can real-time track the position of the center of mass of a moving image. Owing to the vMOS circuit technology, such a processing can be done in several tens nano seconds using a very simple hardware. The characteristic feature of the vMOS integrated circuits is that it accepts analog signals, processing in analog/digital merged hardware computation and yields answers all in digital format. Therefore vMOS circuit technology provides an ideal interfacing between the real world and computer world.