Android端OpenCV运行时权限拒绝错误，无法实现目标识别求助

Hey there! Let's work through your OpenCV on Android issues step by step—since you're new to this, I'll break it down clearly so you can follow along.

Logcat errors are almost always the reason your object recognition isn't working, even if the camera seems to run. Start by capturing the exact error messages from Logcat—here are the most common culprits to look for:

• CameraAccessException: Even though you added the CAMERA permission in your Manifest, Android 6.0+ requires dynamic permission requests. You can't just declare it in XML; you need to ask the user for camera access at runtime before using the camera for recognition.
• OpenCV library loading failures: If you're using static OpenCV libraries, make sure your project includes .so files that match your app's target ABIs (like arm64-v8a or armeabi-v7a). Mismatched ABIs will cause library load errors that break downstream logic.
• Null pointer exceptions in frame processing: If the camera frame you're getting is null, or converting it to an OpenCV Mat fails, your recognition code won't execute at all.

Your provided config is incomplete—here's what you need to add to avoid common issues:

<uses-permission android:name="android.permission.CAMERA"/>
<!-- Optional: Add if using Camera2 API or audio-related features -->
<uses-permission android:name="android.permission.RECORD_AUDIO" android:required="false"/>
<uses-feature android:name="android.hardware.camera" android:required="false"/>
<uses-feature android:name="android.hardware.camera.autofocus" android:required="false"/>
<uses-feature android:name="android.hardware.camera.front" android:required="false"/>

<application
    ...
    android:hardwareAccelerated="false"> <!-- Many OpenCV operations don't work with hardware acceleration -->
    <activity
        ...
        android:screenOrientation="landscape"> <!-- Most OpenCV samples use landscape to avoid frame rotation bugs -->
    </activity>
</application>

Note: Setting android:required="false" means your app can install on devices without a camera, but if your object recognition relies on a camera, set android:required="true" for the camera feature, or add a runtime check to handle camera-less devices.

If the camera runs but recognition doesn't work, the problem is almost always in frame processing:

• Frame format mismatch: Camera previews usually output YUV-formatted frames (like NV21), but most OpenCV recognition algorithms (Haar cascades, YOLO, etc.) need RGB Mat objects. Convert the frame first:

  // Example: Convert NV21 camera frame to RGB Mat
  Mat rgbMat = new Mat();
  Mat yuvMat = new Mat(frameHeight + frameHeight/2, frameWidth, CvType.CV_8UC1);
  yuvMat.put(0, 0, cameraFrameData);
  Imgproc.cvtColor(yuvMat, rgbMat, Imgproc.COLOR_YUV2RGB_NV21);
  

• Invalid recognition model setup: If you're using a pre-trained model (like a Haar cascade), make sure the model file is in your assets folder and loaded correctly. For example, loading a face detection cascade:

  CascadeClassifier faceDetector = new CascadeClassifier();
  InputStream assetStream = getResources().getAssets().open("haarcascade_frontalface_default.xml");
  File cascadeDir = getDir("cascade", Context.MODE_PRIVATE);
  File cascadeFile = new File(cascadeDir, "haarcascade_frontalface_default.xml");
  
  // Copy asset file to internal storage (OpenCV can't load directly from assets)
  FileOutputStream outputStream = new FileOutputStream(cascadeFile);
  byte[] buffer = new byte[4096];
  int bytesRead;
  while ((bytesRead = assetStream.read(buffer)) != -1) {
      outputStream.write(buffer, 0, bytesRead);
  }
  assetStream.close();
  outputStream.close();
  
  // Load the classifier
  faceDetector.load(cascadeFile.getAbsolutePath());
  

• Mismatched frame size: If your model was trained on a specific image size (e.g., 320x240), resize the camera frame to match that size before running recognition:

  Mat resizedFrame = new Mat();
  Imgproc.resize(rgbMat, resizedFrame, new Size(320, 240));
  

You don't need external links—stick to these reliable, accessible resources:

• OpenCV Android Samples: Included in the official OpenCV SDK or the OpenCV GitHub repo. These samples have working code for camera integration, face detection, and object tracking—run them first, then compare your code to see where you're missing steps.
• OpenCV Android Official Documentation: Covers everything from setting up the library in Android Studio to basic image processing and camera frame handling. It's tailored specifically for Android developers.
• Android Official Permission Guides: Focus on dynamic camera permission requests—this is a common pitfall for new Android developers working with camera features.

内容的提问来源于stack exchange，提问作者Taha