Camera崩溃：JNI代码触发Fatal signal 6 (SIGABRT)问题求助

Hey there! Let's work through this crash issue you're hitting when detecting a person and overlaying a shirt image with OpenCV on Android. That Fatal signal 6 (SIGABRT) error is a native-level crash, usually triggered by invalid memory access or a critical OpenCV operation failure. Let's break down the most likely causes and fixes step by step.

First, Understand the SIGABRT Trigger #

SIGABRT in Android's native code almost always points to one of these common issues:

• Null pointer dereference (accessing a Mat or object that's empty/uninitialized)
• Array out-of-bounds errors (trying to access pixels outside the frame/image dimensions)
• Mismatched image formats/sizes between your camera frame and shirt image
• JNI layer data mishandling (incorrectly passing data between Java and C++)
• Memory exhaustion or race conditions in camera processing threads

Step-by-Step Troubleshooting & Fixes #

1. Pinpoint the Exact Crash Location #

First, we need to find which line of code is causing the crash. Since you're a beginner, start with verbose logging in your C++ code to track what's happening right before the crash:

// Add these logs in your detection/overlay logic (include <android/log.h>)
__android_log_print(ANDROID_LOG_DEBUG, "OpenCV_DEBUG", "Camera frame size: %dx%d", frame.cols, frame.rows);
__android_log_print(ANDROID_LOG_DEBUG, "OpenCV_DEBUG", "Detected person: x=%d, y=%d, w=%d, h=%d", rect.x, rect.y, rect.width, rect.height);

// Check if your shirt image is loaded properly
if (shirtMat.empty()) {
    __android_log_print(ANDROID_LOG_ERROR, "OpenCV_ERROR", "Shirt image is EMPTY!");
} else {
    __android_log_print(ANDROID_LOG_DEBUG, "OpenCV_DEBUG", "Shirt image size: %dx%d", shirtMat.cols, shirtMat.rows);
}

If you can use Android Studio's Native Debugger, attach it to your app and let it break on the crash—this will show you the exact call stack and line number causing the issue.

2. Fix Invalid Detection Coordinates #

A super common culprit here is the detected person's rectangle going outside the camera frame bounds. When you try to overlay the shirt on an out-of-bounds area, you'll get an array out-of-bounds crash. Add a boundary check to clamp the rectangle to the frame:

// Clamp the detected rect to fit within the camera frame
int safeX = max(0, rect.x);
int safeY = max(0, rect.y);
int safeWidth = min(rect.width, frame.cols - safeX);
int safeHeight = min(rect.height, frame.rows - safeY);

// Skip overlay if the rect is invalid (e.g., negative size)
if (safeWidth <= 0 || safeHeight <= 0) {
    __android_log_print(ANDROID_LOG_WARN, "OpenCV_WARN", "Invalid person rect—skipping overlay");
    return;
}

3. Ensure Your Shirt Image Loads Correctly #

If your shirt image fails to load (empty Mat), any operation on it will crash. Double-check how you're loading the image:

• If loading from assets (common in Android), use this safe method in JNI:

  // Example: Load shirt.png from assets
  AAsset* asset = AAssetManager_open(manager, "shirt.png", AASSET_MODE_BUFFER);
  if (!asset) {
      __android_log_print(ANDROID_LOG_ERROR, "OpenCV_ERROR", "Couldn't open shirt asset!");
      return;
  }
  long assetSize = AAsset_getLength(asset);
  unsigned char* buffer = (unsigned char*)AAsset_getBuffer(asset);
  Mat shirtMat = imdecode(Mat(1, assetSize, CV_8UC1, buffer), IMREAD_UNCHANGED);
  AAsset_close(asset);
  
  if (shirtMat.empty()) {
      __android_log_print(ANDROID_LOG_ERROR, "OpenCV_ERROR", "Failed to decode shirt image!");
      return;
  }
  

• Make sure the image path/asset name is correct, and the file is a supported format (PNG/JPG).

4. Match Image Formats & Resize Properly #

Camera frames often use RGBA or BGR formats, while your shirt image might have an alpha channel (PNG). Mismatched formats will cause crashes during overlay. Also, resize the shirt to fit the detected person's rectangle:

// Convert shirt to match camera frame format (adjust based on your frame's format)
Mat shirtRgba;
cvtColor(shirtMat, shirtRgba, COLOR_BGRA2RGBA); // Use COLOR_BGR2RGBA if shirt has no alpha

// Resize shirt to fit the safe person rect
Mat resizedShirt;
resize(shirtRgba, resizedShirt, Size(safeWidth, safeHeight));

// Overlay the shirt onto the camera frame
resizedShirt.copyTo(frame(Rect(safeX, safeY, safeWidth, safeHeight)));

5. Fix JNI Memory & Thread Issues #

• Never manually release a Mat that was passed from Java to JNI—Java owns that memory, and freeing it will cause crashes.
• Ensure all OpenCV operations run on the same thread as the camera callback (avoid modifying the same Mat from multiple threads without synchronization).

Final Tips #

Start with the logging step first—it will tell you immediately if your shirt image is empty or if the detection rect is invalid. Those are the two most likely causes for your crash.

内容的提问来源于stack exchange，提问作者Abegail