用FPGA实现视频中物体边缘二值化

       对于高速视频图像中的物体轮廓提取中&＃xff0c;图像的二值化&＃xff0c;采用软件方式&＃xff0c;速度无法满足需要&＃xff0c;因此要采用FPGA方案&＃xff0c;这样可以实现二值化速度和帧率一样&＃xff0c;完全和视频同步。

一、算法原理&＃xff1a;  边缘二值化的算法很多&＃xff0c;象sobel法&＃xff0c;roberts法&＃xff0c;拉普拉斯法&＃xff0c;导数法。为了简单化&＃xff0c;采用导数法。

       视频图像每帧扫描顺序如下图的黑色箭头所示  

 　　　　为了保证二值化和视频同步&＃xff0c;必须要求视频每扫描一个点&＃xff0c;就立刻进行二值化处理&＃xff0c;也就是说每个像素点时钟进行一次。每个点&＃xff08;如上图红色点&＃xff09;进行锐化求导只能和已经出现过的相邻点&＃xff08;如图上图绿色4个点&＃xff09;进行&＃xff0c;当前点的锐化导数值subv为&＃xff1a;

              subv&＃61;max(|v-v10|,|v-v00|,|v-v01|,|v-v02|) ,即和相邻点差值绝对值最大一个作为其锐化值

　　　　  当subv大于规定的值时&＃xff0c;标记该点是1&＃xff0c;否则是0&＃xff0c;实现图像二值化。

二、FPGA结构原理图如下&＃xff1a;

说明&＃xff1a;

     1. CMOS摄像头连接camera接口上&＃xff0c;该接口模块提取视频数据分成2路&＃xff1a;1)保存到双口RAM1中&＃xff0c;以便下一行点锐化用。2&＃xff09;和上一行里的4个数据进行比  较&＃xff0c;得到锐化的二值化数据存放到双口ram2中。

     2. CMOS摄像头同时连接S3c2440(三星arm&＃xff0c;带camera接口)    

     3 。数据读取模块根据TFT时序发生器送来的控制&＃xff0c;选取RAM2或者S3c2440送来的视频数据。通过TFT时序发生器将两个视频数据合并同时显示在屏幕上。这种结构主要目的是把原始图像和处理后的图像同时显示&＃xff0c;以便对比调试。

三、verilog HDL 代码&＃xff1a;

// 定义了一个双口ram存放可存放640*480个二值化像素点

 RAM2PORT cambuf_ram(sharp_caminv,sharp_camreadAdr,ARM_DCLK,sharp_camadr,LVDS_PCLK,1,sharp_camoutv);
//
always &＃64;( negedge LVDS_PCLK) //相机时钟下降沿&＃xff0c;&＃xff08;这里的LVDS把相机所有信号传来&＃xff09;
begin
        if (LVDS_VS&＃61;&＃61;0)   //相机的帧时钟&＃61;0
                begin
                          cam_pcnt&＃61;0;//每行点计数器清
                          cam_hcnt&＃61;0; //每帧行计数器清

                end
        else
                begin
                        if (LVDS_HS&＃61;&＃61;0)  //相机的行时钟
                                begin
                                        cam_pcnt&＃61;0;  //每行点有效点计数器清空
                                        hs1&＃61;0;
                               end
                        else
                                begin
                                        if (hs1&＃61;&＃61;0)
                                        begin
                                                hs1&＃61;1;
                                                cam_hcnt&＃61;cam_hcnt&＃43;1; //每帧行计数器加
                                        end
                                      cam_pcnt&＃61;cam_pcnt&＃43;1;
                                      if ((cam_pcnt<1280)&&(cam_hcnt<480)) //注意每行有效点1280,不是640&＃xff0c;因为YUV格式&＃xff0c;每个点需要传送两个时钟
                                                begin
                                                        camadr&＃61;cam_pcnt[15:1; 
                                                        sharp_camadr&＃61;cam_hcnt*640&＃43;camadr; //二值化双口RAM地址
                                                        
                                                        if(cam_pcnt%2&＃61;&＃61;1)
                                                         begin
                                                               camLastV10<&＃61; camLastV11;  //获取对应邻近点值V10
                                                               camLastV11<&＃61;LVDS_D;        //当前点值v11是当前点
                                                               
                                                               camLastV00<&＃61; camLastV01;//获取对应邻近点值V00
                                                               camLastV01<&＃61; camLastV02;获取对应邻近点值V01
                                                                camLastV02<&＃61;camLastV11buf[camadr&＃43;1]; //获取对应邻近点值V02,注意V02是从上一行的RAM
                                                               camLastV11buf[camadr]<&＃61;LVDS_D;//保存当前点到上一行中
                                                               
                                                               //以下是和相邻的4个点进行比较&＃xff0c;最大差值作为锐化导数值
                                                              subv1&＃61;camLastV00>camLastV11?camLastV00-camLastV11:camLastV11-camLastV00;
                                                              subv2&＃61;camLastV10>camLastV11?camLastV10-camLastV11:camLastV11-camLastV10;
                                                              subv3&＃61;camLastV01>camLastV11?camLastV01-camLastV11:camLastV11-camLastV01;
                                                              subv4&＃61;camLastV02>camLastV11?camLastV02-camLastV11:camLastV11-camLastV02;
                                                              subv&＃61;subv1>subv2?subv1:subv2;
                                                              subv&＃61;subv>subv3?subv:subv3;
                                                              subv&＃61;subv>subv4?subv:subv4;
                                                              sharp_caminv&＃61; (subv>sharpv0)?1:0; //二值化&＃xff0c;结果保存到双口ram中
                                                             
                                                         end
                                                end
                                end
                end
end

四、运行结果照片和视频

      该FPGA软件方案在北京数磊的IEC03工业智能相机运行结果如下&＃xff1a;

     1. 照片

   2.演示视频&＃xff08;录像时用手机录的&＃xff0c;不是很清楚&＃xff0c;请大家包涵了&＃xff09;

     http://v.youku.com/v_show/id_XNDQ1ODg1ODUy.html

      机器视觉产品