Zedboard & Zynq 图像采集 视频开发 (二) FPGA图像采集raw转rgb888

前一篇已经介绍了Zedboard上面搭建图像采集系统的硬件结构，这一篇主要介绍Zynq内部Fpga部分系统设计

下面是在本工程的系统框图

图像采集

OV7725是一款CMOS图像传感器，VGA分辨率，帧率60fps，SCCB协议，与I2C协议通用。这里用HDL编写一个I2C 模块，对ov7725进行初始化，i2c模块是直接采用的crazybingo的设计，这里不再赘述，只是列出ov7725寄存器初始化列表，让其输出Bayer RGB图像数据

0 :     LUT_DATA    =   {8'h1C, 8'h7F}; //Manufacturer ID Byte - High (Read only)
1 :     LUT_DATA    =   {8'h1D, 8'hA2}; //Manufacturer ID Byte - Low (Read only)
//Write Data Index
2   :   LUT_DATA    =   {8'h12, 8'h80}; // BIT[7]-Reset all the Reg
3   :   LUT_DATA    =   {8'h3d, 8'h03}; //DC offset for analog process
4   :   LUT_DATA    =   {8'h15, 8'h02}; //COM10: href/vsync/pclk/data reverse(Vsync H valid)
5   :   LUT_DATA    =   {8'h17, 8'h22}; //VGA:  8'h22;  QVGA:   8'h3f;
6   :   LUT_DATA    =   {8'h18, 8'ha4}; //VGA:  8'ha4;  QVGA:   8'h50;
7   :   LUT_DATA    =   {8'h19, 8'h07}; //VGA:  8'h07;  QVGA:   8'h03;
8   :   LUT_DATA    =   {8'h1a, 8'hf0}; //VGA:  8'hf0;  QVGA:   8'h78;
9   :   LUT_DATA    =   {8'h32, 8'h00}; //Bit[7]:Mirror image edge alignment
10  :   LUT_DATA    =   {8'h29, 8'hA0}; //VGA:  8'hA0;  QVGA:   8'hF0
11  :   LUT_DATA    =   {8'h2C, 8'hF0}; //VGA:  8'hF0;  QVGA:   8'h78
12  :   LUT_DATA    =   {8'h0d, 8'h41}; //Bypass PLL
13  :   LUT_DATA    =   {8'h11, 8'h01}; //CLKRC,Finternal clock = Finput clk*PLL multiplier/[(CLKRC[5:0]+1)*2] = 25MHz*4/[(x+1)*2]
                                        //00: 50fps, 01:25fps, 03:12.5fps   (50Hz Fliter)
14  :   LUT_DATA    =   {8'h12, 8'h03}; //BIT[6]:   0:VGA; 1;QVGA
                                        //BIT[3:2]: 01:RGB565
                                        //VGA:  00:YUV; 01:Processed Bayer RGB; 10:RGB; 11:Bayer RAW; BIT[7]-Reset all the Reg
15  :   LUT_DATA    =   {8'h0c, 8'h90}; //COM3: Bit[6]:Horizontal mirror image ON/OFF, Bit[0]:Color bar; Default:8'h10
//DSP control
16  :   LUT_DATA    =   {8'h42, 8'h7f}; //BLC Blue Channel Target Value, Default: 8'h80
17  :   LUT_DATA    =   {8'h4d, 8'h09}; //BLC Red Channel Target Value, Default: 8'h80
18  :   LUT_DATA    =   {8'h63, 8'hf0}; //AWB Control
19  :   LUT_DATA    =   {8'h64, 8'hff}; //DSP_Ctrl1:
20  :   LUT_DATA    =   {8'h65, 8'h00}; //DSP_Ctrl2:
21  :   LUT_DATA    =   {8'h66, 8'h00}; //{COM3[4](0x0C), DSP_Ctrl3[7]}:00:YUYV;    01:YVYU;    [10:UYVY]   11:VYUY
22  :   LUT_DATA    =   {8'h67, 8'h02}; //DSP_Ctrl4:[1:0]00/01: YUV or RGB; 10: RAW8; 11: RAW10
   //AGC AEC AWB
23  :   LUT_DATA    =   {8'h13, 8'hff};
24  :   LUT_DATA    =   {8'h0f, 8'hc5};
25  :   LUT_DATA    =   {8'h14, 8'h11};
26  :   LUT_DATA    =   {8'h22, 8'h98}; //Banding Filt er Minimum AEC Value; Default: 8'h09
27  :   LUT_DATA    =   {8'h23, 8'h03}; //Banding Filter Maximum Step
28  :   LUT_DATA    =   {8'h24, 8'h40}; //AGC/AEC - Stable Operating Region (Upper Limit)
29  :   LUT_DATA    =   {8'h25, 8'h30}; //AGC/AEC - Stable Operating Region (Lower Limit)
30  :   LUT_DATA    =   {8'h26, 8'ha1}; //AGC/AEC Fast Mode Operating Region
31  :   LUT_DATA    =   {8'h2b, 8'h9e}; //TaiWan: 8'h00:60Hz Filter; Mainland: 8'h9e:50Hz Filter
32  :   LUT_DATA    =   {8'h6b, 8'haa}; //AWB Control 3
33  :   LUT_DATA    =   {8'h13, 8'hff}; //8'hff: AGC AEC AWB Enable; 8'hf0: AGC AEC AWB Disable;
//matrix sharpness brightness contrast UV
34  :   LUT_DATA    =   {8'h90, 8'h0a};
35  :   LUT_DATA    =   {8'h91, 8'h01};
36  :   LUT_DATA    =   {8'h92, 8'h01};
37  :   LUT_DATA    =   {8'h93, 8'h01};
38  :   LUT_DATA    =   {8'h94, 8'h5f};
39  :   LUT_DATA    =   {8'h95, 8'h53};
40  :   LUT_DATA    =   {8'h96, 8'h11};
41  :   LUT_DATA    =   {8'h97, 8'h1a};
42  :   LUT_DATA    =   {8'h98, 8'h3d};
43  :   LUT_DATA    =   {8'h99, 8'h5a};
44  :   LUT_DATA    =   {8'h9a, 8'h1e};
45  :   LUT_DATA    =   {8'h9b, 8'h3f}; //Brightness
46  :   LUT_DATA    =   {8'h9c, 8'h25};
47  :   LUT_DATA    =   {8'h9e, 8'h81};
48  :   LUT_DATA    =   {8'ha6, 8'h06};
49  :   LUT_DATA    =   {8'ha7, 8'h65};
50  :   LUT_DATA    =   {8'ha8, 8'h65};
51  :   LUT_DATA    =   {8'ha9, 8'h80};
52  :   LUT_DATA    =   {8'haa, 8'h80};
//Gamma correction
53  :   LUT_DATA    =   {8'h7e, 8'h0c};
54  :   LUT_DATA    =   {8'h7f, 8'h16}; //
55  :   LUT_DATA    =   {8'h80, 8'h2a};
56  :   LUT_DATA    =   {8'h81, 8'h4e};
57  :   LUT_DATA    =   {8'h82, 8'h61};
58  :   LUT_DATA    =   {8'h83, 8'h6f};
59  :   LUT_DATA    =   {8'h84, 8'h7b};
60  :   LUT_DATA    =   {8'h85, 8'h86};
61  :   LUT_DATA    =   {8'h86, 8'h8e};
62  :   LUT_DATA    =   {8'h87, 8'h97};
63  :   LUT_DATA    =   {8'h88, 8'ha4};
64  :   LUT_DATA    =   {8'h89, 8'haf};
65  :   LUT_DATA    =   {8'h8a, 8'hc5};
66  :   LUT_DATA    =   {8'h8b, 8'hd7};
67  :   LUT_DATA    =   {8'h8c, 8'he8};
68  :   LUT_DATA    =   {8'h8d, 8'h20};
//Others
69  :   LUT_DATA    =   {8'h0e, 8'h65};//night mode auto frame rate control

Bayer8转RGB888

ov7725图像传感器可以输出rgb565，yuv格式的图像数据，但这里用到的是RAW格式图像数据。配置好ov7725以后，摄像头传回raw格式图像数据，每个像素8bit，Bayer像素排列如下

像素总数的一半是G分量，四分之一是R，四分之一是B，这是因为人眼对绿色分量最为敏感，所以G分量占的比重大。将Bayer raw图像转换为RGB图像，最简单的办法就是使用插值法，插值法就是使用周围邻域像素来恢复自身像素值的办法，本设计采用3X3窗口来做双线性插值，举例说明，对于像素点G22，绿色分量就等于自身像素值，红色分量

R22 = (R12 + R32) / 2 蓝色分量 B22 = (B21 + B23) / 2 以此类推

线性插值法简单，而且适合FPGA上面流水线操作，但是也有比较严重的弊端，那就是色彩的恢复没有考虑到各个颜色分量之间的相关性，容易产生伪彩色，同时在图像纹理比较丰富的地方，容易造成图像轮廓失真。

在FPGA上面实现双线性插值法，要对图像进行3行的缓存，这是FPGA图像处理比较基本的东西，也不再多说，只是提一下这件事，一些人在使用Altera FPGA的时候，quartus提供这样一个IP

实际上就是多维的移位寄存器，但是在xilinx的ip中并没有这个ip，只有一维的ram based shift register，不过不要紧，只要将多个一维移位寄存器串接起来就可以实现相同的功能了，当然也可以使用sysgen或者hls自己做一个（个人想法，没有实现过）

做好移位寄存器，就可以生成3X3矩阵，然后线性插值将RAW8转换为RGB888了，具体过程可以参见我的工程：

http://download.csdn.net/detail/zhangyu_eeprom/7726541