The abstract detailed introduction based on the SPCA563B chip’s pattern recognition system’s design concept and the hardware circuit, gives really applies specifically but actually. Carries on the pretreatment to the CMOS gathering image data, then carries on the division and the extraction characteristic to it. And uses certain algorithm to carry on the pattern recognition. The experimental result proved that this system regarding stationary picture’s recognition is very effective.
The key word SPCA563B image knowledge hacks the hand signal recognition
Monolithic integrated circuit SPCA563B is one which the Ling Yang science and technology company promotes uses in the pattern recognition domain IC, not only has the common monolithic integrated circuit’s control function, and has the pattern recognition and handling ability. This pattern recognition system is carries on the data processing and the control take it as the core. This system has realized the function includes: Object color and shape recognition, character recognition as well as hand signal recognition and so on. Using these recognition’s result, may develop application systems and so on game, interest teaching, intelligent toy, writing input as well as intelligent recognition.
This article through the introduction overall system as well as the simple hand signal recognition which realizes on this system, makes everybody to understand this system well in the pattern recognition and the writing recognition aspect formidable function and the usability.
1 system hardware construction
Like chart l shows. The pattern recognition hardware system mainly by the SPCA563B monolithic integrated circuit, the USB controller, the image sensor, the memory controller and the speech output controller and so on is composed.
1.1 SPCA563B monolithic integrated circuits
The SPCA563B monolithic integrated circuit’s hardware characteristic is as follows: Provides 320×240 8 CMOS image sensor connection; Has automatic is in vain balanced and obtains the parameter control function; Has the image color adjustment and the image Gamma adjustment function; Has 7 kind of color filtration function; Has good reduces the noise the filter; Can examine in a picture to 21 goal object size and the position, and its result depositing in inserting in CPU internal RAM; 16 μnSP CPU essences, in inlay 1K×16 position RAM and 32K×16 position ROM; Has exterior RAM which and ROM connection may choose; Has the USBl.1 connection; 3.3V voltage power supply; 128/64 foot LQFP seal (128 feet have parallel interface which may choose, 48 feet do not have). This system uses 48 foot’s LQFP seals, its hardware circuit as shown in Figure 2.
1.2 USB controllers
In SPCA563B inlays USBl.1 the controller, may carry on the correspondence using the USB connection with PC, unifies the PC end software to realize ISP and the online debugging routine, in addition may come using USB for the system power supply. Its electric circuit as shown in Figure 3.
1.3 image sensors
This system uses SPCA3010The CMOS image sensor, the picture output is QVGA(320×240) size RGB form. SPCA563B controls SPCA3010A through the I2S connection the CMOS image sensor. This CMOS image sensor supports stand the by province electricity pattern, its electric circuit as shown in Figure 4.
1.4 memory controllers
Outside expands SPR4096The Flash memory, mainly uses for the pronunciation data which the memory system needs. SPR4096A FJash has the following characteristic: 512K×8 position storage space; In inlays 4K×8 position SRAM; Exterior CPU may visits Flash/SRAM through the serial interface or 8 bit parallel connection; The I/O connection’s voltage range is 2.25~3.6 V, and supports stancl by the province electricity pattern. In the data reserves quite small situation, may reduce system’s cost greatly with SPR40396A. Electric circuit as shown in Figure 5.
1.5 speech outputs control
The speech output controller’s leading role is enlarges the SPCA563B two group audio frequency output through SPY0030, broadcasts by the speaker. Its hardware circuit as shown in Figure 6.
2 system software designs
This pattern recognition system’s recognition function mainly realizes through the software system. In order to enhance the procedure the probability, will be the future this system function expansion reservation space, the entire software system will use a minute module, the method write program in administrative levels, will use the assembly language to realize to the register reads/writes as well as to the interrupt the control, and will provide the nimble connection for the upper formation development. The upper formation uses the C language compilation, through transfers the first floor function to come to carry on the operation indirectly to the hardware source. Such write program causes the overall system structure to be clear, the procedure probability and the extendibility strengthen, moreover takes the hardware source to be few, and further develops the reservation full space for system function’s expansion.
The entire software system by the master routine, the pattern recognition functional module subroutine, the interrupt servicing subroutine and the speech output subroutine and so on is composed. The master routine is responsible for overall system’s coordinated and the control work, realizes the different function through the transfer different subroutine. Master routine flow as shown in Figure 7.
For the prominent system’s characteristic, below comes the key description pattern recognition principle through the hand signal recognition’s example. The hand signal recognition contains two steps: The hand shape’s contour tracing, withdraws its outline the correlation data; Hand signal recognition, in previous step obtains in the data foundation carries on the analysis judgment to obtain the result.
2.1 contour tracing algorithm realizes
Contour tracing’s goal is obtains the image the exterior outline characteristic (including goal object peripheral point coordinate and direction), prepares for the image shape analysis. SPCA563B DSP has a series of color processing in view of the skin color, such system may facilitate the image to transform as the skin color and the non-skin color two value image. Through carries on the data analysis to the interim in the SRAM two value image, extraction shape edge contour.
The contour tracing algorithm step is as follows:
①According to, the progressive scanning image from left to right, seeks for the first non-white picture element to take boundary initial station A[0] from top to bottom, records A[0] coordinate (A[0] .x,A[0].y). A[0] has smallest line and the row value boundary point. Defines scanning direction variable dir again, this variable uses in recording one in step along the preceding boundary point to the current boundary point travel direction. Its initialization value is A[0].dir=7; Scanning direction definition as shown in Figure 8.
The supposition current coordinate is (x, y), then its 8 neighborhood coordinate is as follows:
②According to counter clockwise search current picture element 3×3 the neighborhood, its outset search direction hypothesis is as follows:
If A[n-1].dir is an odd number, then takes (A[n-1] .dir 7) rood 8;
If A[n-1].dir is an even number, then takes (A[n-1] .dir 6) mod 8.
First which searches in 3×3 the neighborhood with current picture element same picture element then for new boundary point A[n], simultaneously the transaction logging searches from previous to boundary point A[n] direction variable A[n].dir, record new boundary point coordinate (A[n] .x,A[n].y).
③If boundary point A[n] is equal to first boundary point A[0], namely (A[n] .x==A[O].x&&A[n].y==A[0].y). Stop search. Conclusion track. Otherwise redundant step②.
By boundary point A[O], A[1], A[2],…, A[n] constitution’s boundary then for the boundary which must track. In algorithm step①The function foot discovers the first boundary point, the step③The function discovers all boundary points. As shown in Figure 9, left side is the primitive image, right side is with this algorithm simulation result.
2.2 hand signal recognition algorithm realizes
Our main purpose is discovers stretches out finger’s integer. Starts with the line scan method through the line from O→1 or 1→O the black and white picture element change number of times determined that stretches out finger’s integer, but this algorithm regarding inclined certain angle finger number judgment existence certain question, therefore transfers uses “withdraws a fingertip algorithm”.
Withdraws a fingertip method to focus in searches the fingertip spot from the hand signal outline, a fingertip point correspondence in a finger, acts according to finally obtains a fingertip integer to determine the finger number. Is the judgment the fingertip: The analysis shape, nearby finger’s two basically is parallel. Has pressed the counter clockwise advance, each process fingertip, the hand shape outline’s trend has the reverse change. As a result of hand shape outline edge existence certain burr, therefore may use the multi-spot direction the average to reduce the burr the influence. Here takes 36 directions the averages, the expression is Average_dix[i]. Whether after fingertip’s judgment condition 1:126<=|Average_dir[i-1] Average_dir[i 1]|<=162. (explained: Two directions are opposite, its direction value difference 4,3.5×36=126,4.5×36=162.)
The analysis shape, between two fingers edge contour’s trend also satisfies judgment condition l again, must therefore increase the judgment condition. As shown in Figure 10, when along a direction search edge contour, moves toward after the fingertip the gap moves certainly toward with the finger between opposite. Therefore, must the supposition determination condition 2:
Is_FingerVertex(dir0, dirl, dir2);
This function is uses for through 3 spot directions to judge this 3 spots continuously whether is the anti-clockwise trend. Because searches the edge contour along the counter clockwise, therefore adds on under the determination condition 1 establishment’s premise Is_FingerVertex(Average_dir[i-1], Average_dir[i], Average_dir[i 1]) may judge whether is the fingertip, thus determines finger’s number. Whether the smooth 3 spots are the anti-clockwise trends, makes concrete may construct a Round-robin queue in the procedure to realize. The test result indicated that this algorithm is stable.
Conclusion
SPCA563B internal inserting function formidable image color processing DSP, makes the pattern recognition system’s master control chip with it to be very convenient, quickly. Simultaneously this system has the speech output prompt facility, the operation user-friendly, and may carries on the correspondence through the USBl.1 connection and PC; Coordinates the PC end the debugging aids, but may also reduce pattern recognition function development greatly the cycle.
