snapshot mode is provided。 Both the column  and row skip modes of the CMOS sensor reduce image size without reducing field of view。 The features of THDB-

     1 1 2 2 1 2

F (,)F * (,)

ej (1x0 2 y0 )

(3)

D5M are controlled by customized Hardware Description Language (HDL), such as gain, frame rate, frame size, exposure [9], [10]。

The resolution of the output image is the full width and

After have the inverse FFT, a correlation function   can

be acquired

⎡ * ⎤

height of the field of view in the default settings。 There

d F 1 ⎢F1 (1,2 )F2 (1,2 ) ⎥F 1ej (1 x0 2 y0 )

are two methods to reduce the output resolution: Skipping

and    Binning。    Row    and    column    skip    modes use ⎣

subsampling to reduce the output resolution without reducing field of view [11], [12]。 The sensor also has row and column binning modes, which can reduce the  impact

⎦⎥

(4)

Figure 4。   The Consequence of the test for detecting the special region。

In Equation (4), the  equation  d  is  a  two- dimensional function。 The overlapped area corresponds to coordinate (x0 , y0 ) can be obtained through the max value of [17]。

Although the performance of phase correlation algorithm is verified through the simulation test in Matlab, the FFT for the whole image which more than 800×480 pixels is too complicated to achieve in embedded system。

Hence, we optimize the algorithm by select special region pixels。 To detect the special region, five hundreds pair of test were set and the consequence is shown in Fig。 4。 Since the position of two image sensor is settled, the coordinate of the overlap region is aggregate between 190 to 197 from x-coordinate and 0 to 7 from y-coordinate。 In this case, the special region contains 64 pixels and the FFT for special region can fulfill in FPGA embedded system。

B。 RAW To RGB Module

To acquire more real information of the view, the best solution is use three CMOS chips to obtain the red, green and blue independently。 Then combines the three   colors

filter mosaic which is a color filter array (CFA) for arranging RGB color filters on a square grid of photo sensors [18]。 This unprocessed image form named raw format and therefore is not ready to be used with a bitmap graphics editor or printed。

The raw format output of Bayer-filter cameras is referred to as a Bayer pattern image。 Since each pixel is filtered to record only one of three colors, the data from each pixel cannot fully determine its own color [19]。 The structure of RAW to RGB Module is presented in Fig。 5。 The data of RAW file firstly passes the Line_Buffer_0 and Line_Buffer_1 to save first two rows data in these buffers。 Then set the B0_1 with first pixel of row one, as shown in B, and set the B0_2 with first pixel of row two, as shown in G2。 The same as B1_1 and B1_2, they are set by the second pixel of first two rows, as shown in G2 and

R。 After that, all of B0_1, B0_2, B1_1 and B1_2 send to MUX logical to gain the RGB data through maintain the value of B0_1 and B1_2 as the value of blue color and red color, and get the sum of B0_2 and B1_1, two values of green, as the value of green color。

C。RGB To YUV Module