This paper analyzes the present s i t u a t i o n concerning bearing surface appearance defect inspection. Also, a demonstration i s given of the i n t r i n s i c handicap o f the most uidely used analogus thresholding technique f o r inproving inspection r e l i a b i l i t y . Thereafter, ue propose a test machine v i s i o n inspection system. The system u t i l i z e s an SSPDA (self-scanning photodiode array) as the l i g h t sensor, a special mechanism f o r u n r o l l i n g the r o l l e r surface and a PC/AT f o r image processing.The performance of the system i s controlled by a UC-280 single board conputer. For the image processing, three novel methods uere used: 1) average based shading correction, 2) backgrwnd model for p o l l u t i o n erasing, and 3) a local i t e r a t i v e operator for defect and background clustering. These algorithms make the defect segmentation very reliable, even i n the preswre of o i l pollution. For different precision rollers, the system can easily be adapted by only a l t e r i n g the input parameters. Test results show that the system i s capable of inspecting most o f the surface defects such as cracks, pits, rust or scratches u i t h high r e l i a b i l i t y .The m i n i m crack uidth uhich can be detected i s 25m a t present. The significance of t h i s work i s i n presenting a machine v i s i o n inspection system u i t h high r e l i a b i l i t y and processing speed that i s able t o be inplemnted on carmercially available harduare image processing architecture and solve a practical problem i n bearing inspection. 1-IYTlKDUCTlOYThe inspection of surface appearance defect of r o l l e r bearings i s s t i l l basically unsolved. An automated system could inprove the quality and production of bearings.The major d i f f i c u l t y i s that there i s s t i l l no effective method uhich can inspect the r o l l e r u i t h both high r e l i a b i l i t y and high efficiency. Most of the presently used equipment requires that the r o l l e r rotate a t high speed, and the inspection takes place a t each individual point being s c a d . This characteristic r e s t r i c t s the substantial inprovement of r e l i a b i l i t y , as surface p a r t i c l e adhesion or time-based pulse noise c a m t be discriminated from the real defect i n 1-D time base axis.According t o recent survey L1.2.3.41, Rost a u t m t i c bearing inspection machines produce too many nisdetections t o be accepted for practical application.Therefore, finding a n e w uay t o solve t h i s problem becomes essential.Uith the developnent o f signal processing technology, as well as the advent of high capacity microconputers, attention of inspection has been directed t o the use of image processing techniques and tremendous achievements have been made i n many f i e l d s such as electronics, lunber, nuclear and t e x t i l e industries C5.6.7.81. Y i t h t h i s method, the inspected surface i s captured as an image and a computer i s used t o analyze the defective pattern. Because the image i ...

No abstract

Tracking a moving target of interest at a high resolution with a dynamically designated network camera while ensuring complete-coverage of the area under surveillance of a video camera network can be a challenging task. Game theory can be applied to the situation, treating individual cameras as players and area coverage as utility. Camera collaboration is needed when one camera handoff the job of tracking a moving target of interest to another. By taking into account coverage shared by two cameras as secondary utility and giving precedence to border-covering camera directions, we can improve the performance of collaborative tracking, reducing complete-coverage failures as well as their associated uncovered blocks between handoffs. The cost of camera adjustment involved in a handoff can also be reduced this way.

In light of the history of LDPC codes and relevant research advances in recent years, this paper probes into the encoding and decoding techniques related to this capacity-approaching error-correction technology. Besides the general expression as an equation, LDPC codes can also be examined with a Tanner graph. The encoding of LDPC codes comprises two tasks: construct a sparse parity-check matrix, and generate codewords with the matrix. The decoding of LDPC codes can be divided into three phases: initialization, message update, and validation. With a conventional model of communication systems, common decoding algorithms of LDPC codes are scrutinized.In particular, the sum-product algorithm is analyzed in an elaborate fashion. The logarithmic sumproduct algorithm and the min-sum algorithm are two important variations of the sum-product algorithm. The logarithmic sum-product algorithm reduces multiplication to addition by introducing logarithmic likelihood ratio while the latter simplifies computation at the cost of precision.

No abstract