Object monitoring and identification for robot

مقدم أطروحة جامعية

Abdul, Husam Kazim

الجامعة

الجامعة التكنولوجية

الكلية

-

القسم الأكاديمي

قسم الهندسة الكهروميكانيكية

دولة الجامعة

العراق

الدرجة العلمية

دكتوراه

تاريخ الدرجة العلمية

2004

الملخص الإنجليزي

In this research, an investigation of robot and camera configuration (eye-in-hand) was presented to improve its performance.

This configuration detects the object within the field of the view.

Recognition process is applied to enable object matching with a predefined model.

The real Cartesian coordinates of the monitored object are recovered relative to the image plane data.

The recovery process includes the identifying of the X and Y coordinates of the object in addition to object depth.

Four main processes are investigated to improve the performance of the eye-in-hand configuration.

These are : Determining of the orientation of the monitored object with respect to the reference object (object orientation).

I Matching the monitored object with a predefined or reference object, (object matching Adjusting the camera scene to assure the capturing of the complete object image (object complementing Recovery of 'object coordinates and depth (coordinates recovery).

In the first part, two methods were adopted and programmed to determine the orientation of the monitored object with respect to-the.

Reference object.

These methods are based on cross-correlation: and principal axes techniques.

The results of both methods were investigated and analyzed using many representative testing samples.

Errors measure (delta errors) was proposed and-used to evaluate results of object 'matching.

While in the second part, three matching methodologies were developed and programmed to achieve the recognition process.

These methods are maximum radius.

Mean radius and four 'radii methods.

The results of these methods showed validity.

However the four radii method gave better results to.

Overcome scaling and rotation effects.

The developed method based on cross-correlation presents a beneficial recognition-capability.

The zoom out effect was investigated to evaluate, its drawback on object matehirm and the determination of object orientation A proposed algorithm was prepared and programmed to achieve the object complement process that assures the monitored object to be within the field of view.

The proposed algorithm is based on 16 crossing cases of the image boundaries.

To activate the operation of the proposed algorithm, a neural system was developed, where it consists of two neural networks.

The first one is the main network while the other works as a feedback unit to eliminate the effect of the close loop movement.

The neural networks are trained using backpropagation method, and are examined and showed acceptable results A developed method for estimating the object depth was prepared and programmed based on image and physical areas.

The developed method shows good efficiency to recover the object depth.

To accelerate and generalize the method, a proposed neural network was designed and trained using samples which were obtained rom building the mathematical model of the experimental data.

This neural network was aPphed and showed satisfactory results.

the developed work shows an improvement of eye-in-hand overall capability to Perform intelligently robot tasks.

التخصصات الرئيسية

الهندسة الكهربائية

الموضوعات

• أنظمة التحكم
• الروبوتات

• APA
• MLA
• AMA

لغة النص

الإنجليزية

نوع البيانات

رسائل جامعية

رقم السجل

BIM-305999