Abstract: This article introduced one kind based on the double TMS320VC5416 processor’s real-time visual search track processing system, elaborated in detail this system’s hardware design thought that the parallel connection united plants the track algorithm example to narrate based on the DSP visual search and the track processing system software design general flow. This system is one kind of fine imagery processing platform, has the good versatility, may use for to realize many kinds of imagery processing algorithm.
Key word: Imagery processing digital signal processor search track
1. introduction
The television image tracking device is one kind has the simple intelligent image follow-up device, it can in the quite complex background, according to the standard television system, withdraw and separate in the field of view by the field the movement goal, the extraction object brightness and the structural feature, the positioning of the target center to the turnplate axis of the eye’s position and the angle of pitch error. This tracking angle error message after transformation delivers turnplate’s server, actuates the turnplate movement to guarantee that the goal which tracks is been in image-forming sensors’ field of view center throughout, realizes to the goal automatic tracking.
Because needs to chase the field (20ms) to process in the field of view the data, therefore processes the data quantity is big, the algorithm order of complexity is high, the traditional processor cannot satisfy the speed request generally, this system chooses American TI Corporation TMS320C5416 signal processor is a core, has realized in the real-time gathering field of view image data and completes the corresponding imagery processing algorithm operation the duty. The TMS320C5416 basic frequency may reach 160MHz, the internal total storage space for 128M×16bit, is a section of high performance low power loss general digital signal processing chip. This system can process 50/second, image resolution adjustable standard television image signal.
2. system hardware diagram
Figure 1 system hardware overall construction diagram
As shown in Figure 1 the system to use the double DSP CPLD skeleton, the system has two groups inputs, a group meets the digital video signal input, other group inputs the PAL service pattern from the camera the video signal. In double DSP DSP, is responsible to process the track algorithm as well as with the superior machine correspondence primarily, other is responsible from DSP real-time to produce simulates the Gauss noise to use to examine under each kind of noise condition to track the algorithm the effect. Between two DSP carries on the correspondence through sharing pair of mouth RAM or HPI. The regulation image data on the scene saves to image SRAM, advocates the DSP counter regulation on the scene to read the image data from image pair of mouth RAM to DSP, the field regulation starts when advocates DSP to start to carry on the imagery processing algorithm, will advocate the result which as well as the correlation data in the next counter regulation DSP will process reads in graphical display pair of mouth RAM simultaneously to start from the SRAM read-in next data, the DSP processing to complete later in the sequence circuit and under the video frequency compound circuit coordination will process the result to demonstrate on the monitoring device, will complete the real-time imagery processing task.
2.1 image gathering module
The image gathering module’s major function is gains in the input video signal gradation data and the synchronized clock, it is the following processing datum. The system uses the sync split and the phase lock technique design, uses the discrete component. Realizes is specifically the signal comes out after CCD divides into two groups, behind pass sync split synchronization separator LM1881, output compound synchronization HS, field locking VS takes electric circuit’s control signal, another pass 篏 position and the cocurrent restore, then the enlargement, adjusts the image signal in a/D switch’s reference voltage scope. Carries on the phase-lock frequency multiplication to the good synchronized signal then to obtain the picture element clock signal, the phase-locked loop chip uses 74HC4046. The input video signal provides after the phase-locked loop phase-lock output system picture element clock to a/D converter use, obtains the digital image data.
2.2 sequence circuit module
The succession module mainly (Xilinx Corporation’s 95288XL) realizes by piece of CPLD, including the phase-lock counting, the standard video frequency good field signal production, the DSP external connection memory interface selects patches or strips of land as worth saving for seed the read-write signal production as well as the partial memory address production, realizes the figure signal and the string transforms, the user from to define I/O and so on.
2.3 communication interface module
This system with receives the PC transmission to an asynchronous serial port the debug command, and to PC returns operation result.
5416 provide the serial port is one kind of synchronized serial interface, does not support the general asynchronous receiver/transmitter (UART) standard, this system uses MAXIM Corporation’s MAX3100 chip to realize the synchronized serial port to the asynchronous serial port transformation. 5416 use the synchronized signal which FSR and FSX take each time transmit, FSX takes MAX3100 the gating signal. Synchronized receive clock CLKR and synchronized transmission clock CLKX uses the internal clock source in this system, and takes CLKX MAX3100 the synchronized clock. In the system uses MAX3100 the receive interrupt to take DSP the external interrupt signal, informs DSP data preparation, may start to receive.
2.4 memory visit module
The image buffer uses single mouth large capacity SRAM, may save the entire field image. The regulation receive on the scene the data which gathers from the gathering module. Counter regulation DSP on the scene will wait processing the data to read in internal data area processing by SRAM.
Graphic data pair of mouth RAM uses in the character which, the graph saves uses in demonstrating and so on. In a field of view determination of fix corresponds in memory’s one unit (byte) (bit). DSP is going to demonstrate the character or the graph in the memory cell which stipulated by the lattice form write store, the memory the read-out must demonstrate the character or the graph lattice signal under the system sequential control. 8 bitmap figurate numbers according to pass through CPLD and - the string switching circuit turn the serial signal iteration to provide the demonstration on the simulated video signal.
3. system software design
This system may take digital image processing the general platform, the processing many kinds of image processing program, the software design is nimble.
Advocates DSP to use to complete the search track algorithm, the search, the track algorithm has many kinds, presently as regularly explains the image track processing software design take the classics related track algorithm the general flow. If the need, may increase the different track algorithm.
The related track uses the image similarity measure method, seeks for the optimum matching subsegment in the image the work, may select the algorithm including the normalized mutual correlation (NCC) and the mean deviation accumulation and (MAD) and so on. Because the related operational data process load is big, to be advantageous real-time realizes, this system uses the smallest absolute error accumulation and related matching algorithm. This method when calculates two image f1, f2 between them the similar measure carries on according to the equation below:
And f1, f2 express the template and the search sector image subsegment image separately, after computation all search sector chart subsegment and template image absolute error accumulation and C, determination smallest C correspondence subsegment position namely optimum matching spot.
Regarding 32×32 template and 64×64 the search sector image, each image determined merely the optimum matching position the computation number of times is, in addition work and so on template revision and policy-making judgment, each data computation load is very big. For example, when the instruction cycle is 6ns, the operation time approximately is 7ms. In order to reduce the computation load, after may use the image resolution to take the lead the thick, fine method. The image resolution reduces one time, the computation load reduces close 15 times. In the actual system, the template thick sampling match track execution time approximately is 1ms, may satisfy the system timely request.
The practice proved that in carries on the sequence image track process, if comes the current image’s optimum matching position’s image as the template to carry on purely the next image the match, the track result is very easy one to occur the sudden change image influence to deviate the correct position. Therefore, must consider that acts according to the old template and the current image optimum matching position match (gathers moderately) formulates the appropriate new template, is equal to matches the track process to carry on an instruction, achieves the quite good track effect. This system design one kind of template weighting amends a plan, namely
And: A is the original template image content; B For this match optimum matching position sub-image content;
After M is the revision, template image content; W is the weighting factor (in basis frame and frame is related confidence selection)
In track process, because the background is complex, will also possibly have situations and so on partial mask, this system took the anti-mask measure to eliminate the partial mask influence to a certain extent. Its basis is in has in the mask situation, the optimum matching position’s absolute error accumulation and has not had mask time will be much bigger than, will carry on the mask appraisal in the match process to the match result to be as follows:
1st, first determined that the mask area threshold is the mask part occupies the template area the percentage threshold, surpasses the threshold to think that is covered;
2nd, then the basis optimum matching position’s average picture element absolute error for the reference determination mask picture element’s gradation threshold, the real-time search sector correspondence template region’s picture element grey level is bigger than the gradation threshold, then thought that this picture element is covered;
3rd, the computation current optimum matching position image and the template image absolute error surpasses the mask picture element gradation threshold integer S, then judges S whether to surpass the mask the area threshold, if surpasses, then thought that the goal is covered, sets the goal to cover the flag bit, adjusts the initialization match result storehouse routine, the template adjusts the cyclical counter homing stop adjustment template, then current match result with front certain match result ensemble average replace; Otherwise thought that has not been covered, the clear goal covers the flag bit.
The system software realizes the process is specifically: First carries on the initialization to the system. In the initialization subroutine defines a condition character SysStatus control procedure the flow, for example when SysStatus=0 is the search condition, when SysStatus=1 is related when track algorithm SysStatus=2 is the contrast gradient track condition and so on. After the initialization had ended, the system enters the circulation waiting status, each field inquires the SysStatus condition, the waiting status change. The tracking object template’s initial capture may choose in the search condition completes by the target acquisition algorithm, may also by complete artificially. Take the artificial capture as the example, the user in the superior PC machine field of view the selected target region, then and searches the target location coordinate the region size information to transmit through the RS232 serial port the condition evaluation which receives for the DSP, DSP basis gives global variable SysStatus. Then the value which instructed according to the SysStatus parameter enters the different tracking mode.
Specifically related track algorithm flow following shown in Figure 2:
Figure 2 related match flow chart
After a data processing completes, a result pass serial port feeds back to the superior machine, a pass graph memory, in the target location draws the window and demonstrates on the monitoring device. Then starts to carry on next processing.
Completes the real-time Gauss noise from DSP the production, the noise superimposes to the gathering input picture may use to use for to examine the different track algorithm under the different intensity Gauss noise actual result.
The algorithm for becomes the Gauss stochastic noise by the even noise through the mapping hypergene, has, the high accuracy merit high speed. Functional block diagram as shown in Figure 3: The left half part for the even noise production part, uses the modulus to take -odd law, depends upon existing k in (- x, x) in the uniform distribution, mutually the independent random number seed x(1), ….. x(k), the unceasing production new random number, and will output outward. The right half part uses the left part the value of exports is the offset address, in addition the right table’s base address production reflects the firing table address, after table look-up, obtains this address result output obtained the result Gauss annoyance value which requests, then enters the next circulation.
Figure 3 noise production schematic diagram
Mainly completes the noise from DSP the production work, if the system does not need to simulate the Gauss noise, this DSP may also serve as other use according to the actual need, increased system’s extendibility.
4. concluding remark
This system used integration rates and so on DSP, CPLD on the hardware to be high, the power loss low advanced component, caused the system circuit to have the volume to be small, the weight was light, the power loss was low, the reliable higher merit, was one kind of fine imagery processing platform. Succeeded on this system has realized based on the direction code target acquisition algorithm as well as related, the contrast gradient and so on many kinds of track algorithms, the experiment proved that this system satisfied the timely request completely, could stabilize, precise and the fast search, the tracking object.
Reference:
[1] Mr. Pan An, Zhang Kai, Yan Zhaochun and so on. TMS320 series chip in television tracking device application. Systems engineering and electronic technology, 1994.3:pp1~6
[2] TMS320VC5416 Digital Signal Processor Data Sheet. Literature number:SPRS095H. Texas Instruments Incorporated (TI), March 1999 - Revised December 2001
[3] Opens grandly, Cao Tieyong arranges the .DSP chip the principle and the development application (the second edition). Beijing: Electronics industry publishing house, 2000.9
