Pattern recognition system using optical analogue processing

Pattern recognition system using optical analogue processing

Abstracts Introductory talk: the future of lightwave sensing Hirokazu Matsumoto National Umezono Research 1- l-4, 16th Meeting Laboratory of Metro...

147KB Sizes 0 Downloads 30 Views

Abstracts Introductory talk: the future of lightwave sensing Hirokazu Matsumoto National Umezono

Research

1- l-4,

16th Meeting

Laboratory of Metrology Tsukuba, Ibaraki 305, Japan

on Lightwave

(NRLM),

Sensing Technology

Lightwave technology has the possibility objects or phenomena remotely and to Here, some items concerning two or sensing are discussed for industrial and

to measure three-dimensional process information quickly. three-dimensional lightwave scientific applications.

Shin Yamamato, Keiichi Yamada, Tomoaki Nakano

16th Meeting

on Lightwave

16th Meeting

This paper describes the technical trend, current research topics and technical issues of machine vision in the automotive industries. Machine vision technologies have been widely applied and much research has been done in many fields, including factory automation, and an intelligent transportation system. In recent years, more developments have begun for on-board driving environmental recognition and driver’s face recognition. To develop such a system a number of techniques such as monocular vision, binocular vision and optical flow, have been tried, but no reliable method has been developed to extract the object from a complicated background. To solve this problem, it is necessary to develop an imaging method that is robust to brightness changes, and to develop an algorithm for recognition.

N. Yasuda, H. Hosokawa, T. Yamashita

16th Meeting

on Lightwave

Sensing Technology

Akira Hirano, Masahide Tsujishita Fundamental Research Laboratories, Osaka Gas Co, Ltd, 619-9, Torishima, Konohana-ku, Osaka 554, Japan Sensing Technology

This paper describes an analytical technique for prompt NO formation and measurement of flame temperature by planar laser-induced fluorescence (PLIF). For NO generation, the concentration mappings of CH (10 ppm), CN (sub ppm) and NH (< 100 ppb) (which are involved in prompt NO formation) and NO itself, with a spatial resolution of 200 urn in flame cross-sections, have been obtained. This demonstrates that the PLIF technique is effective for prompt NO

& Laser Technology Vol29 No I 1997

In the detection of moving space objects, such as space debris and satellites, the images of objects to be detected are often small and dim. A new detection method for these objects, in a sequence of digital images, is presented. In order to reduce the noise and clutter, preprocessing techniques are used. The processed image is then segmented so as to obtain some candidate targets using a threshold estimated according to its histogram and possible target size. After processing several successive frames, the target trace may be searched and identified. The simulation result is used to demonstrate the utility of the algorithm.

Pattern recognition system using optical analogue processing Haruyoshi Toyoda

16th Meeting

The imaging diagnositics of flames by means of planar laserinduced fluorescence

on Lightwave

Sensing Technology

on Lightwave

Photonics

K.K.,

Sensing Technology

Shimokaiinji,

In recent years, a high performance sensing technique has been required to meet the demands of the high quality and the diversification of industrial products. New sensing techniques, which enable not only on/off detection but also detection of other useful information about an object, are required to be developed in the field of factory automation. We have developed several shining surface sensors that detect gloss by using a new polarizing beam-splitter. In this paper, the design of the optical system, the property of the new polarizing beam-splitter, which is indispensable for this sensor, and the effects on several applications are discussed.

16th Meeting

on Lightwave

Central Research Laboratory, Hamamatsu 5000 Hirakuchi, Hamakita 434, Japan

High performance sensors using polarization

Corporation,

Ruiming Li, Tadishi Aruga

Nagakute,

Sensing Technology

Central R&D Lab, OMRON Nagaokakyo, Kyoto 617, Japan

A new image processing method to detect a dim, moving object Communications Research Laboratory, Ministry of Posts and Telecommunications, 4-2-1, Nukui-Kitamachi, Koganei-shi, 184, Tokyo, Japan

Machine vision for automotive applications Toyota Central R&D Labs, Inc, 41-l Yokomichi, Nagakute-cho, Aichi 480- 11, Japan

generation diagnostics. For temperature, two-dimensional measurement of the NQ rotational temperature has been demonstrated using PLIF. This method has the advantage of avoiding the limit of the measuring area due to the molecular distribution zone by seeding NO into secondary air or fuel, as well as giving us the opportunity to measure the flame temperature above 2100 K. These results are acquired under low concentration and low quantum yield conditions (methane-air flame), showing that this analytical technique is applicable to studies on practical combustors.

The optical anaiogue processing technique has a massively parallel computing capability for pattern recognition and image processing. One of the key devices in the system is a spatial light modulator (SLM) which has the ability to control the optical amplitude and phase twodimensionally. By using a phase-only modulating SLM developed by our group, we have demonstrated a real-time (video-frame-rate) optical correlator and have applied it to fingerprint identification and velocity measurement. It has been conftrmed that the optical analogue processing can give real processing capability to the practical processing systems.

Terminal attractor optical associative memory for pattern recognition Xin Lin, Junji Ohtsubo Faculty of Engineering, Hamamatsu, 432, Japan 16th Meeting

Shizuoka

on Lightwave

University,

Johoku

3-5-1,

Sensing Technology

To alleviate the spurious state in the Hopfield neural network, the concept of terminal attractors (TA) has been introduced. In this paper, we apply the TA model to an associative memory neural network for pattern recognition and compare it with the conventional Hopfield model. The computer simulations for the pattern recognition have been performed by using 6x6 neurons and three stored patterns in the network. For the feasibility of optical implementation of the TA model, a neural network with 16 neuron and three stored patterns is tested in the experiment. These results indicate that the TA model can reduce spurious states in the Hopfield neural network and the recalling capability can be much improved.

optics

xiii