角蜂鸟实例教程 > Android > 手绘识别

cll826

手绘识别

本文主要给大家介绍下在Android平台下使用角蜂鸟实现手绘识别功能

准备工作

• 配置环境等详情请参照Hello 2018里的快速开始，此处不具体阐述
• 下载物体检测所需模型graph_sg文件，在您Android Studio中，当前module下新建assets包，将下载的模型文件复制到该目录下
• 因工程需要处理图像，所以使用了javacv库，可从GitHub自行下载或点击链接从示例工程中libs下拷贝到自己工程
• 将class_list_chn.txt文件拷贝到Android设备的存储空间下，用于345种物体的比对
  

具体实现

• 将graph文件传输到角蜂鸟里
  
  

1. int status = allocateGraphByAssets("graph_sg");

复制代码

• 读取物体分类文件
  

1. try {
   
2. BufferedReader bufferedReader = new BufferedReader(new FileReader(Environment.getExternalStorageDirectory().getAbsolutePath() + "/hs/class_list_chn.txt"));
   
3. for (int i = 0; i < 345; i++) {
   
4. String line = bufferedReader.readLine();
   
5. if (line != null) {
   
6. String[] strings = line.split(" ");
   
7. mObjectNames[i] = strings[0];
   
8. }
   
9. }
   
10. bufferedReader.close();
    
11. } catch (FileNotFoundException e) {
    
12. e.printStackTrace();
    
13. Log.e("SketchGuessThread", "FileNotFoundException");
    
14. } catch (IOException e) {
    
15. e.printStackTrace();
    
16. Log.e("SketchGuessThread", "IOException");
    
17. }

复制代码

• 获取角蜂鸟内置的摄像头图像
  

1. byte[] bytes = deviceGetImage();

复制代码

注意：开发者也可以根据自己需求选择自己外部摄像头数据。

• 处理图像，并将处理好的tensor传给角蜂鸟
  
  

1. opencv_core.IplImage bgrImage = opencv_core.IplImage.create(FRAME_W, FRAME_H, opencv_core.IPL_DEPTH_8U, 3);
   
2. bgrImage.getByteBuffer().put(bytes_frame);
   
3. int sg_weight = (int) (FRAME_W * roi_ratio);
   
4. //crop
   
5. opencv_core.CvRect cvRect = opencv_core.cvRect((int) (FRAME_W * (0.5 - roi_ratio / 2)), (int) (FRAME_H * 0.5 - sg_weight / 2), sg_weight, sg_weight);
   
6. cvSetImageROI(bgrImage, cvRect);
   
7. opencv_core.IplImage cropped = cvCreateImage(cvGetSize(bgrImage), bgrImage.depth(), bgrImage.nChannels());
   
8. cvCopy(bgrImage, cropped);
   
9. //canny
   
10. opencv_core.IplImage image_canny = opencv_core.IplImage.create(sg_weight, sg_weight, opencv_core.IPL_DEPTH_8U, 1);
    
11. cvCanny(cropped, image_canny, 120, 45);
    
12. //dilate
    
13. opencv_core.IplImage image_dilate = opencv_core.IplImage.create(sg_weight, sg_weight, opencv_core.IPL_DEPTH_8U, 1);
    
14. //kernel = np.ones((4,4),np.uint8)
    
15. opencv_core.IplConvKernel iplConvKernel = cvCreateStructuringElementEx(4, 4, 0, 0, CV_SHAPE_RECT);
    
16. cvDilate(image_canny, image_dilate, iplConvKernel, 1);
    
17. opencv_core.IplImage image_dilate_rgba = opencv_core.IplImage.create(sg_weight, sg_weight, opencv_core.IPL_DEPTH_8U, 4);
    
18. cvCvtColor(image_dilate, image_dilate_rgba, CV_GRAY2RGBA);
    
19. //resize
    
20. opencv_core.IplImage image_load = opencv_core.IplImage.create(28, 28, opencv_core.IPL_DEPTH_8U, 4);
    
21. cvResize(image_dilate_rgba, image_load);
    
22. //把该image_dilate转化成float类型传送给角蜂鸟
    
23. Bitmap bitmap_tensor = IplImageToBitmap(image_load);
    
24. Message message1 = new Message();
    
25. message1.arg1 = 2;
    
26. message1.obj = bitmap_tensor;
    
27. mHandler.sendMessage(message1);
    
28. int[] pixels = new int[28 * 28];
    
29. bitmap_tensor.getPixels(pixels, 0, 28, 0, 0, 28, 28);
    
30. float[] floats = new float[28 * 28 * 3];
    
31. for (int i = 0; i < 28 * 28; i++) {
    
32. floats[i] = Color.red(pixels[i]) * 0.007843f - 1;
    
33. floats[3 * i + 1] = Color.green(pixels[i]) * 0.007843f - 1;
    
34. floats[3 * i + 2] = Color.blue(pixels[i]) * 0.007843f - 1;
    
35. }
    
36. int status_tensor = loadTensor(floats, floats.length, 0);

复制代码

• 获取返回的处理结果
  

1. float[] result = mHsApi.getResult(0);

复制代码

• 处理result返回值,排列取出最大的5个数对应的idex.
  

1. public String[] sortMax5(float[] result) {
   
2. 
   
3.    HashMap<Integer, Float> integerFloatHashMap = new HashMap<>();
   
4.    String[] object_names = new String[5];
   
5.    for (int i = 0; i < result.length; i++) {
   
6.        integerFloatHashMap.put(i, result[i]);
   
7.    }
   
8.    Arrays.sort(result);
   
9.    for (int i = 0; i < result.length; i++) {
   
10.        if (integerFloatHashMap.get(i) == result[result.length - 1]) {
    
11.            object_names[0] = mObjectNames[i] + " " + result[result.length - 1];
    
12.        }
    
13.        if (integerFloatHashMap.get(i) == result[result.length - 2]) {
    
14.            object_names[1] = mObjectNames[i] + " " + result[result.length - 2];
    
15.        }
    
16.        if (integerFloatHashMap.get(i) == result[result.length - 3]) {
    
17.            object_names[2] = mObjectNames[i] + " " + result[result.length - 3];
    
18.        }
    
19.        if (integerFloatHashMap.get(i) == result[result.length - 4]) {
    
20.            object_names[3] = mObjectNames[i] + " " + result[result.length - 4];
    
21.        }
    
22.        if (integerFloatHashMap.get(i) == result[result.length - 5]) {
    
23.            object_names[4] = mObjectNames[i] + " " + result[result.length - 5];
    
24.        }
    
25.    }
    
26.    return object_names;
    
27. }

复制代码

• 结果显示在view上,效果如图：
  

友好提醒：因Android设备基本都是USB2.0,所以不建议使用1080P的图像，传输比较耗时，会有卡顿感，可以使用360P的图像，铺满屏幕即可

具体代码可去GitHub下载，地址如下 SungemSDK-AndroidExamples

产品购买链接：https://www.cirmall.com/shop/?c=products&a=view&id=5113