YUV module which can concert bitmap image to YUV image for compress the view image, and Fast Fourier Transform module which implement the optimal phase correlation algorithm。 The Ethernet interface fulfill with a new structure of embedded system without RTOS and TCP/IP protocol stack to develop the speed of network transmission。 The whole system was achieved in the DE2-70 FPGA embedded framework。 The main architecture of the camera is shown in Fig。 1。

III。IMAGING OF CMOS SENSOR

A。 Optical Structure of Panoramic Perspective

To obtain panoramic perspective of scene, panoramic mosaic camera contains two optical acquisition systems combined with two CMOS sensors。 The optical structure of panoramic perspective is presented in Fig。 2。 The filed of   view   of   each   optical   subsystem   is ,   and the

overlapped  area  is 。  The  optical  axes  of  subsystem

are L1 and L2 , which are parallel to the center line L 。

The distance between L1 and L2 is h 。 The images of region received by two image sensors are same。 The key function for this camera is automatically detect the coordinate of the overlapped area and use algorithm to fusion the two images into one panoramic image。

Figure 2。   Optical Structure of Panoramic Perspective。

of aliasing introduced by the use of skip modes。 Both 2× and 4× binning modes are supported [13]。 Rows and columns can be binned independently。 As shown by Fig。 3。

To drive the sensor, the settings should be configured and the configuration instructions of the CMOS sensor can be sent through I2C module。 According to the datasheet of the sensor, the driver of CMOS sensor is designed in Verilog [14]。

2X Skip Mode 2X Bin & Skip Mode

Figure 3。   The Two binning mode of the sensor。

IV。IMAGE MOSAIC

A。Optimal Mosaic Algorithm

The basic principle of mosaic methodology is phase correlation algorithm which is a nonlinear method based on Fast Fourier Transform。 This algorithm  only relate with the information of cross-power spectrum。 As a result, it is undependable with the specific content of image [15], [16]。 Beside, phase correlation algorithm does not rely on the value of gray image。 So as to, it is very robust to obstruction and has advantages to porting in embedded system。

If we wish to match images f1 (x, y) and f2 (x, y) , and only translational radiation exist between f1 and f2 。 It means

B。CMOS Sensor

f1 (x, y) 

f2 (x x0 , y y0 )

(1)

THDB-D5M CMOS sensor from Terasic Corporation is used in this panoramic mosaic camera。 It contains  two

Substituting to FFT features, the result of conversion is

j (1x0 2 y0 )

different  operation  modes:  full  frame  mode  and frame

F1 (1,2 ) e F2 (1,2 )

(2)

transfer mode。 The array size of the THDBD5M is 2592×1944 [7], [8]。 There is only one readout register  in

In   this   equation,

F1 (1,2 )

and

F2 (1,2 )  is

the sensor, located at the top of the image area。 The register is driven independently by two phase clocks。 This CMOS sensor has high frame rate and low dark current。 Moreover,  to  meet  the  demand  of  taking  frames,   the

corresponding FFT result of F1  and F2 。 The   cross-power

spectrum is

F (,)F * (,)