DEV Community

A Bit Above Bytes
A Bit Above Bytes

Posted on

Stop Taking Blurry Scans: How I Built Frame Stability Detection for Android with CameraX

#android #kotlin #tutorial #mobile

Every document scanner app has the same problem: users tap the capture button while their hand is still moving, and the scan comes out blurry.

I spent two weeks solving this while building a production Android scanner app (21K lines of Kotlin). Here's the approach that actually works.

The Problem

CameraX gives you a camera preview, but it has no concept of "is the frame stable enough to capture?" You need to build that yourself.

The Solution: RMS-Based Frame Stability

Instead of complex optical flow or gyroscope fusion, I used a simple but effective approach: compare consecutive frames using Root Mean Square (RMS) difference.

class FrameStabilityAnalyzer : ImageAnalysis.Analyzer {
    private var lastFrameBytes: ByteArray? = null
    private var isStable = false
    private val STABILITY_THRESHOLD = 5.0 // Tune this value

    override fun analyze(imageProxy: ImageProxy) {
        val currentBytes = imageProxy.planes[0].buffer.toByteArray()

        lastFrameBytes?.let { previous ->
            val rms = calculateRMS(previous, currentBytes)
            isStable = rms < STABILITY_THRESHOLD
        }

        lastFrameBytes = currentBytes
        imageProxy.close()
    }

    private fun calculateRMS(a: ByteArray, b: ByteArray): Double {
        val size = minOf(a.size, b.size)
        var sumSquares = 0.0
        for (i in 0 until size) {
            val diff = (a[i].toInt() and 0xFF) - (b[i].toInt() and 0xFF)
            sumSquares += diff * diff
        }
        return sqrt(sumSquares / size)
    }
}
Enter fullscreen mode Exit fullscreen mode

How It Works

  1. Grab the luminance plane from each camera frame (Y plane in YUV)
  2. Compare it to the previous frame pixel by pixel
  3. Calculate the RMS difference -- low values mean the camera isn't moving
  4. Gate the capture button -- only allow capture when isStable == true

The Key Insight

The threshold value (STABILITY_THRESHOLD = 5.0) is critical. Too low and the user can never capture. Too high and you still get blurry scans.

I found 5.0 works well for document scanning where you want sharp text. For general photos, you might go up to 8-10.

Wiring It Up with CameraX 

val stabilityAnalyzer = FrameStabilityAnalyzer()

val imageAnalysis = ImageAnalysis.Builder()
    .setBackpressureStrategy(ImageAnalysis.STRATEGY_KEEP_ONLY_LATEST)
    .build()
    .also { it.setAnalyzer(executor, stabilityAnalyzer) }

// In your UI, observe stabilityAnalyzer.isStable
// Show a green border when stable, red when shaking
Enter fullscreen mode Exit fullscreen mode

Pro Tips

  • Only analyze a downsampled version of the frame for performance. You don't need full resolution to detect motion.
  • Add a debounce -- require 3 consecutive stable frames before allowing capture. This prevents false positives.
  • Show visual feedback -- users need to know WHY the button is grayed out. A subtle "hold steady" indicator works better than nothing.

Results

After implementing this, the percentage of blurry scans in my app dropped by roughly 80%. Users don't even notice the gating -- they naturally hold still for a fraction of a second, and the capture fires automatically.

This is one of those details that separates a good scanner from a great one. If you're building anything with CameraX that involves document capture, frame stability detection is worth the effort.

I'm currently building this as part of a full production scanner app template. If you're interested in skipping the months of work, check out my profile for more details.

Top comments (0)