pulse-monitor/detection.go

794 lines
24 KiB
Go

package main
import (
"fmt"
"image"
"image/color"
"math"
"sort"
"gocv.io/x/gocv"
)
// DEBUG flag is declared in types.go
// Band represents a vertical band of contours
type Band struct {
minX int
maxX int
minY int
maxY int
}
// DetectBands finds and groups contours into vertical bands
// Only returns bands with height >= minHeight pixels
func DetectBands(binary gocv.Mat, minHeight int) []Band {
type Box struct {
minX int
maxX int
minY int
maxY int
}
// Remove thin vertical lines (UI borders) using morphological opening with horizontal kernel
// This breaks the connection between rows caused by vertical UI elements
kernel := gocv.GetStructuringElement(gocv.MorphRect, image.Pt(15, 1)) // Wide horizontal kernel
cleaned := gocv.NewMat()
gocv.MorphologyEx(binary, &cleaned, gocv.MorphOpen, kernel)
kernel.Close()
defer cleaned.Close()
// Find contours on the cleaned image
contours := gocv.FindContours(cleaned, gocv.RetrievalExternal, gocv.ChainApproxSimple)
// Get all bounding boxes, filtering out tall thin UI elements (borders)
boxes := []Box{}
frameHeight := binary.Rows()
for i := 0; i < contours.Size(); i++ {
rect := gocv.BoundingRect(contours.At(i))
width := rect.Max.X - rect.Min.X
height := rect.Max.Y - rect.Min.Y
// Skip tall thin contours (aspect ratio > 8) that span most of frame height
// These are likely UI borders, not content
if width > 0 && height > frameHeight*6/10 {
aspectRatio := float64(height) / float64(width)
if aspectRatio > 8 {
if DEBUG {
fmt.Printf("DEBUG: Skipping border contour: %dx%d (aspect=%.1f)\n", width, height, aspectRatio)
}
continue
}
}
boxes = append(boxes, Box{
minX: rect.Min.X,
maxX: rect.Max.X,
minY: rect.Min.Y,
maxY: rect.Max.Y,
})
}
// Sort boxes by minY
sort.Slice(boxes, func(i, j int) bool {
return boxes[i].minY < boxes[j].minY
})
// Group overlapping/adjacent boxes into vertical bands
bands := []Band{}
for _, box := range boxes {
// Skip boxes shorter than minHeight
height := box.maxY - box.minY
if height < minHeight {
continue
}
if len(bands) == 0 {
// First band
bands = append(bands, Band{
minX: box.minX,
maxX: box.maxX,
minY: box.minY,
maxY: box.maxY,
})
} else {
lastBand := &bands[len(bands)-1]
// Check if this box overlaps or is close to last band (within 5px)
if box.minY <= lastBand.maxY+5 {
// Merge: extend the band in both X and Y
if box.minX < lastBand.minX {
lastBand.minX = box.minX
}
if box.maxX > lastBand.maxX {
lastBand.maxX = box.maxX
}
if box.maxY > lastBand.maxY {
lastBand.maxY = box.maxY
}
} else {
// New band
bands = append(bands, Band{
minX: box.minX,
maxX: box.maxX,
minY: box.minY,
maxY: box.maxY,
})
}
}
}
return bands
}
// findBaseline analyzes the baseline in the graph band
// Returns left point, right point, width, number of sections, and error
func findBaseline(binary gocv.Mat, graphBand Band) (*image.Point, *image.Point, int, int, error) {
// Extract just the graph band region
graphRegion := binary.Region(image.Rect(0, graphBand.minY, binary.Cols(), graphBand.maxY+1))
// Find all contours in graph region
contours := gocv.FindContours(graphRegion, gocv.RetrievalExternal, gocv.ChainApproxSimple)
// Get sections (contours) with width >= 5px
type Section struct {
minX int
maxX int
minY int
maxY int
width int
height int
}
sections := []Section{}
allContours := []Section{} // For debug visualization
for i := 0; i < contours.Size(); i++ {
rect := gocv.BoundingRect(contours.At(i))
width := rect.Max.X - rect.Min.X
height := rect.Max.Y - rect.Min.Y
// Collect ALL contours >= 50px high for visualization
if height >= 50 {
allContours = append(allContours, Section{
minX: rect.Min.X,
maxX: rect.Max.X,
minY: rect.Min.Y + graphBand.minY,
maxY: rect.Max.Y + graphBand.minY,
width: width,
height: height,
})
}
// Filter: width >= 5px AND height >= 50px for baseline detection
// (lowered from 80px to handle smaller waveform sections)
if width >= 5 && height >= 50 {
sections = append(sections, Section{
minX: rect.Min.X,
maxX: rect.Max.X,
minY: rect.Min.Y + graphBand.minY, // Adjust to full image coordinates
maxY: rect.Max.Y + graphBand.minY,
width: width,
height: height,
})
}
}
// DEBUG: Visualize ALL contours >= 50px high
if DEBUG {
viz := gocv.NewMat()
gocv.CvtColor(binary, &viz, gocv.ColorGrayToBGR)
// Draw graph band boundaries
gocv.Line(&viz, image.Pt(0, graphBand.minY), image.Pt(viz.Cols(), graphBand.minY),
color.RGBA{255, 0, 255, 255}, 2)
gocv.Line(&viz, image.Pt(0, graphBand.maxY), image.Pt(viz.Cols(), graphBand.maxY),
color.RGBA{255, 0, 255, 255}, 2)
// Draw ALL contours >= 50px high in yellow
for i, c := range allContours {
rect := image.Rect(c.minX, c.minY, c.maxX, c.maxY)
gocv.Rectangle(&viz, rect, color.RGBA{0, 255, 255, 255}, 2)
label := fmt.Sprintf("#%d: %dx%d", i+1, c.width, c.height)
gocv.PutText(&viz, label, image.Pt(c.minX, c.minY-5),
gocv.FontHersheyPlain, 1.0, color.RGBA{0, 255, 255, 255}, 1)
}
// Highlight sections that passed the 80px height filter in green
for i, s := range sections {
rect := image.Rect(s.minX, s.minY, s.maxX, s.maxY)
gocv.Rectangle(&viz, rect, color.RGBA{0, 255, 0, 255}, 3)
label := fmt.Sprintf("BASELINE#%d", i+1)
gocv.PutText(&viz, label, image.Pt(s.minX, s.maxY+20),
gocv.FontHersheyPlain, 1.2, color.RGBA{0, 255, 0, 255}, 2)
}
// Add summary text
summary := fmt.Sprintf("Contours H>=50px: %d | Baseline sections: %d", len(allContours), len(sections))
gocv.PutText(&viz, summary, image.Pt(10, 30),
gocv.FontHersheyDuplex, 1.0, color.RGBA{255, 255, 0, 255}, 2)
gocv.IMWrite("test_output/debug_all_contours.png", viz)
viz.Close()
fmt.Println("💾 Saved all contours visualization to test_output/debug_all_contours.png")
}
// Check if we have exactly 2 sections
if len(sections) != 2 {
return nil, nil, 0, len(sections), fmt.Errorf("found %d sections (need exactly 2)", len(sections))
}
// Sort sections by X position (left to right)
sort.Slice(sections, func(i, j int) bool {
return sections[i].minX < sections[j].minX
})
// Horizontal width (left edge of left section to right edge of right section)
horizontalWidth := sections[1].maxX - sections[0].minX
// Fixed offset from edges for sampling points
// Needs to be large enough to avoid curved edges due to tilt
offset := 10
// Find rotation points
// Left section: lowest white pixel offset from left edge
leftSection := sections[0]
leftX := leftSection.minX + offset
leftPoint := findLowestWhitePixel(binary, leftX, graphBand.minY, graphBand.maxY)
// Right section: lowest white pixel offset from right edge
rightSection := sections[1]
rightX := rightSection.maxX - offset
rightPoint := findLowestWhitePixel(binary, rightX, graphBand.minY, graphBand.maxY)
if leftPoint == nil || rightPoint == nil {
return nil, nil, 0, len(sections), fmt.Errorf("could not find rotation endpoints")
}
return leftPoint, rightPoint, horizontalWidth, len(sections), nil
}
// findLowestWhitePixel finds the lowest (max Y) white pixel at given X within Y range
func findLowestWhitePixel(binary gocv.Mat, x int, minY int, maxY int) *image.Point {
lowestY := -1
for y := minY; y <= maxY; y++ {
if binary.GetUCharAt(y, x) == 255 {
lowestY = y
}
}
if lowestY == -1 {
return nil
}
return &image.Point{X: x, Y: lowestY}
}
// DetectionResult contains rotation and width detection results
type DetectionResult struct {
Rotation float64
Width int
ScaleFactor float64 // Scale factor to apply (860 / width)
SpO2 image.Rectangle // SpO2 display area in SCALED, ROTATED coordinates
HR image.Rectangle // HR display area in SCALED, ROTATED coordinates
Success bool
}
// PreprocessFrame takes a raw frame and prepares it for detection
// Returns the processed binary image
func PreprocessFrame(frame gocv.Mat) gocv.Mat {
// Crop top 68 pixels (timestamp)
cropped := frame.Region(image.Rect(0, 68, frame.Cols(), frame.Rows()))
// Rotate 90° clockwise
rotated := gocv.NewMat()
gocv.Rotate(cropped, &rotated, gocv.Rotate90Clockwise)
// Convert to grayscale
gray := gocv.NewMat()
gocv.CvtColor(rotated, &gray, gocv.ColorBGRToGray)
rotated.Close()
// Threshold to binary (matching main pipeline)
binary := gocv.NewMat()
gocv.Threshold(gray, &binary, 180, 255, gocv.ThresholdBinary)
gray.Close()
return binary
}
// DetectRotationAndWidth analyzes a binary frame and returns rotation angle, width, and display areas
// Works on its own internal copy - does NOT modify the input frame
// Returns coordinates in TRANSFORMED (scaled to 860px, rotated) coordinate space
func DetectRotationAndWidth(binary gocv.Mat) DetectionResult {
// Clone the input so we don't modify it
workingCopy := binary.Clone()
defer workingCopy.Close()
// ========== PHASE 1: Detect rotation & scale from baseline ==========
// Detect bands (using 40px minimum height for original scale - lowered for stream2 which is lower res)
bands := DetectBands(workingCopy, 40)
// Save debug image showing detected bands (only in DEBUG mode)
if DEBUG {
viz := gocv.NewMat()
gocv.CvtColor(workingCopy, &viz, gocv.ColorGrayToBGR)
colors := []color.RGBA{
{255, 0, 0, 0}, // Red
{0, 255, 0, 0}, // Green
{0, 0, 255, 0}, // Blue
{255, 255, 0, 0}, // Yellow
{255, 0, 255, 0}, // Magenta
{0, 255, 255, 0}, // Cyan
}
for i, band := range bands {
c := colors[i%len(colors)]
gocv.Rectangle(&viz, image.Rect(band.minX, band.minY, band.maxX, band.maxY), c, 2)
gocv.PutText(&viz, fmt.Sprintf("Band %d: h=%d", i, band.maxY-band.minY),
image.Pt(band.minX, band.minY-5), gocv.FontHersheyPlain, 1.0, c, 1)
}
gocv.PutText(&viz, fmt.Sprintf("Total bands: %d (need >=2)", len(bands)),
image.Pt(10, 30), gocv.FontHersheyPlain, 1.5, color.RGBA{255, 255, 255, 0}, 2)
gocv.IMWrite("test_output/debug_bands.png", viz)
viz.Close()
}
// Check if we have any content visible
if len(bands) < 1 {
if DEBUG {
fmt.Printf("DEBUG: Detection failed - no bands found\n")
}
return DetectionResult{Success: false}
}
// For rotation detection, we ideally want a waveform band
// If available, use band 0 as graph band; otherwise skip rotation detection
var graphBand Band
var hasGraphBand bool
if len(bands) >= 1 {
graphBand = bands[0] // Use first band for rotation detection if available
hasGraphBand = true
if DEBUG {
fmt.Printf("DEBUG: Using band 0 as graph band (h=%d)\n", graphBand.maxY-graphBand.minY)
}
}
// Find baseline for rotation detection (only if we have a usable graph band)
var leftPoint, rightPoint *image.Point
var width, sections int
var err error
if hasGraphBand {
leftPoint, rightPoint, width, sections, err = findBaseline(workingCopy, graphBand)
}
// Variables for rotation and scale
var rotation float64
var scaleFactor float64
if !hasGraphBand || err != nil || sections != 2 {
// Baseline detection failed or not available - use fallback (assume no rotation, scale=1)
if DEBUG {
if !hasGraphBand {
fmt.Printf("DEBUG: No graph band - using fallback rotation=0, scale=1\n")
} else if err != nil {
fmt.Printf("DEBUG: Baseline failed (%v) - using fallback\n", err)
} else {
fmt.Printf("DEBUG: Baseline found %d sections - using fallback\n", sections)
}
}
rotation = 0.0
scaleFactor = 1.0 // No scaling for fixed layout approach
width = workingCopy.Cols()
// Create dummy points for visualization
midY := workingCopy.Rows() / 2
leftPoint = &image.Point{X: 0, Y: midY}
rightPoint = &image.Point{X: workingCopy.Cols(), Y: midY}
} else {
// Baseline detection succeeded - calculate rotation and scale
deltaY := float64(rightPoint.Y - leftPoint.Y)
deltaX := float64(rightPoint.X - leftPoint.X)
rotation = math.Atan2(deltaY, deltaX) * 180 / math.Pi
scaleFactor = 860.0 / float64(width)
}
// DEBUG: Visualize the baseline points
if DEBUG {
viz := gocv.NewMat()
gocv.CvtColor(workingCopy, &viz, gocv.ColorGrayToBGR)
// Find the two baseline sections for visualization
graphRegion := workingCopy.Region(image.Rect(0, graphBand.minY, workingCopy.Cols(), graphBand.maxY+1))
contours := gocv.FindContours(graphRegion, gocv.RetrievalExternal, gocv.ChainApproxSimple)
type Section struct {
minX int
maxX int
minY int
maxY int
}
sections := []Section{}
for i := 0; i < contours.Size(); i++ {
rect := gocv.BoundingRect(contours.At(i))
width := rect.Max.X - rect.Min.X
height := rect.Max.Y - rect.Min.Y
if width >= 5 && height >= 80 {
sections = append(sections, Section{
minX: rect.Min.X,
maxX: rect.Max.X,
minY: rect.Min.Y + graphBand.minY,
maxY: rect.Max.Y + graphBand.minY,
})
}
}
sort.Slice(sections, func(i, j int) bool {
return sections[i].minX < sections[j].minX
})
// Draw baseline sections in cyan
for i, sec := range sections {
rect := image.Rect(sec.minX, sec.minY, sec.maxX, sec.maxY)
gocv.Rectangle(&viz, rect, color.RGBA{0, 255, 255, 255}, 2)
label := fmt.Sprintf("Sec%d: %dpx", i+1, sec.maxX-sec.minX)
gocv.PutText(&viz, label, image.Pt(sec.minX, sec.minY-5),
gocv.FontHersheyPlain, 1.0, color.RGBA{0, 255, 255, 255}, 1)
}
// Draw left point (green circle)
gocv.Circle(&viz, *leftPoint, 10, color.RGBA{0, 255, 0, 255}, -1)
gocv.PutText(&viz, "L", image.Pt(leftPoint.X-5, leftPoint.Y-15),
gocv.FontHersheyPlain, 1.5, color.RGBA{0, 255, 0, 255}, 2)
// Draw right point (red circle)
gocv.Circle(&viz, *rightPoint, 10, color.RGBA{0, 0, 255, 255}, -1)
gocv.PutText(&viz, "R", image.Pt(rightPoint.X-5, rightPoint.Y-15),
gocv.FontHersheyPlain, 1.5, color.RGBA{0, 0, 255, 255}, 2)
// Draw line between baseline points (yellow)
gocv.Line(&viz, *leftPoint, *rightPoint, color.RGBA{0, 255, 255, 255}, 3)
// Draw graph band boundaries (magenta)
gocv.Line(&viz, image.Pt(0, graphBand.minY), image.Pt(viz.Cols(), graphBand.minY), color.RGBA{255, 0, 255, 255}, 2)
gocv.Line(&viz, image.Pt(0, graphBand.maxY), image.Pt(viz.Cols(), graphBand.maxY), color.RGBA{255, 0, 255, 255}, 2)
gocv.PutText(&viz, "Graph Band", image.Pt(10, graphBand.minY+20),
gocv.FontHersheyPlain, 1.2, color.RGBA{255, 0, 255, 255}, 1)
// Add measurements text overlay at top
widthText := fmt.Sprintf("Baseline Width: %dpx", width)
gocv.PutText(&viz, widthText, image.Pt(10, 30),
gocv.FontHersheyDuplex, 1.0, color.RGBA{255, 255, 0, 255}, 2)
rotText := fmt.Sprintf("Rotation: %.3f deg", rotation)
gocv.PutText(&viz, rotText, image.Pt(10, 60),
gocv.FontHersheyDuplex, 1.0, color.RGBA{255, 255, 0, 255}, 2)
scaleText := fmt.Sprintf("Scale: %.3f (->860px)", scaleFactor)
gocv.PutText(&viz, scaleText, image.Pt(10, 90),
gocv.FontHersheyDuplex, 1.0, color.RGBA{255, 255, 0, 255}, 2)
gocv.IMWrite("test_output/debug_baseline_detection.png", viz)
viz.Close()
fmt.Println("💾 Saved baseline detection visualization to test_output/debug_baseline_detection.png")
}
if DEBUG {
fmt.Printf("DEBUG: Baseline points: Left(%d,%d) Right(%d,%d)\n",
leftPoint.X, leftPoint.Y, rightPoint.X, rightPoint.Y)
fmt.Printf("DEBUG: Delta Y=%d, Delta X=%d, Rotation=%.3f°\n",
rightPoint.Y-leftPoint.Y, rightPoint.X-leftPoint.X, rotation)
fmt.Printf("DEBUG: Width=%dpx, Scale factor=%.3f (baseline will become 860px)\n",
width, scaleFactor)
}
// ========== PHASE 2: Find the main digit band for SpO2/HR ==========
//
// For this pulse-ox display, use FIXED proportions since the layout is consistent
// and dynamic band detection is unreliable due to vertical UI elements
//
// After 90° CW rotation, the frame layout is approximately:
// - Y 0-35%: Status bar, bell icon
// - Y 35-45%: Waveform display
// - Y 45-60%: Main SpO2/HR digits (what we want!)
// - Y 60-75%: Secondary readings
frameHeight := workingCopy.Rows()
frameWidth := workingCopy.Cols()
// Use fixed proportions for the digit band
digitBand := Band{
minX: int(float64(frameWidth) * 0.05), // Slight margin from left
maxX: int(float64(frameWidth) * 0.75), // Avoid the vertical progress bar on right
minY: int(float64(frameHeight) * 0.45),
maxY: int(float64(frameHeight) * 0.58),
}
if DEBUG {
fmt.Printf("DEBUG: Using fixed digit band: Y[%d-%d], X[%d-%d] (frame %dx%d)\n",
digitBand.minY, digitBand.maxY, digitBand.minX, digitBand.maxX,
frameWidth, frameHeight)
}
// Split digit band: left half = SpO2, right half = HR
bandWidth := digitBand.maxX - digitBand.minX
midX := digitBand.minX + bandWidth/2
// SpO2 is left half, HR is right half
spo2Rect := image.Rect(
digitBand.minX,
digitBand.minY,
midX,
digitBand.maxY,
)
hrRect := image.Rect(
midX,
digitBand.minY,
digitBand.maxX,
digitBand.maxY,
)
if DEBUG {
fmt.Printf("DEBUG: SpO2 display area: X[%d-%d] Y[%d-%d]\n",
spo2Rect.Min.X, spo2Rect.Max.X, spo2Rect.Min.Y, spo2Rect.Max.Y)
fmt.Printf("DEBUG: HR display area: X[%d-%d] Y[%d-%d]\n",
hrRect.Min.X, hrRect.Max.X, hrRect.Min.Y, hrRect.Max.Y)
// Visualize final display boxes
viz2 := gocv.NewMat()
gocv.CvtColor(workingCopy, &viz2, gocv.ColorGrayToBGR)
// Draw SpO2 box in red
gocv.Rectangle(&viz2, spo2Rect, color.RGBA{255, 0, 0, 255}, 3)
gocv.PutText(&viz2, "SpO2", image.Pt(spo2Rect.Min.X, spo2Rect.Min.Y-10),
gocv.FontHersheyDuplex, 1.2, color.RGBA{255, 0, 0, 255}, 2)
// Draw HR box in cyan
gocv.Rectangle(&viz2, hrRect, color.RGBA{0, 255, 255, 255}, 3)
gocv.PutText(&viz2, "HR", image.Pt(hrRect.Min.X, hrRect.Min.Y-10),
gocv.FontHersheyDuplex, 1.2, color.RGBA{0, 255, 255, 255}, 2)
gocv.IMWrite("test_output/debug_final_boxes.png", viz2)
viz2.Close()
fmt.Println("💾 Saved final display boxes visualization to test_output/debug_final_boxes.png")
}
return DetectionResult{
Rotation: rotation,
Width: width,
ScaleFactor: scaleFactor,
SpO2: spo2Rect,
HR: hrRect,
Success: true,
}
}
// RotateImage rotates an image by the given angle in degrees
// The angle is the detected rotation that needs to be corrected
func RotateImage(img gocv.Mat, angleDegrees float64) gocv.Mat {
// Get image center
center := image.Point{
X: img.Cols() / 2,
Y: img.Rows() / 2,
}
// Apply rotation to correct the tilt
// NOTE: If image rotates wrong direction, change to: -angleDegrees
rotMat := gocv.GetRotationMatrix2D(center, angleDegrees, 1.0)
defer rotMat.Close()
// Apply rotation
rotated := gocv.NewMat()
gocv.WarpAffine(img, &rotated, rotMat, image.Point{X: img.Cols(), Y: img.Rows()})
if DEBUG {
fmt.Printf("DEBUG: Applied rotation: %.3f°\n", angleDegrees)
}
return rotated
}
// ExtractBandRegion extracts a specific band region from an image
func ExtractBandRegion(img gocv.Mat, band Band) gocv.Mat {
// Create a region of interest (ROI) for the band
rect := image.Rect(0, band.minY, img.Cols(), band.maxY)
region := img.Region(rect)
// Clone the region to create an independent Mat
extracted := gocv.NewMat()
region.CopyTo(&extracted)
if DEBUG {
fmt.Printf("DEBUG: Extracted band region [%d-%d], size: %dx%d\n",
band.minY, band.maxY, extracted.Cols(), extracted.Rows())
}
return extracted
}
// FindDigitBoxes finds individual digit contours within a region
// Returns bounding rectangles for digits larger than minSize
func FindDigitBoxes(digitRegion gocv.Mat, minSize int) []image.Rectangle {
contours := gocv.FindContours(digitRegion, gocv.RetrievalExternal, gocv.ChainApproxSimple)
var digitBoxes []image.Rectangle
for i := 0; i < contours.Size(); i++ {
rect := gocv.BoundingRect(contours.At(i))
height := rect.Max.Y - rect.Min.Y
// Filter by minimum size - height only
if height >= minSize {
digitBoxes = append(digitBoxes, rect)
}
}
// Sort boxes by X position (left to right)
sort.Slice(digitBoxes, func(i, j int) bool {
return digitBoxes[i].Min.X < digitBoxes[j].Min.X
})
if DEBUG {
fmt.Printf("DEBUG: Found %d digit contours (H>=%dpx)\n", len(digitBoxes), minSize)
// Visualize found boxes
visualization := gocv.NewMat()
defer visualization.Close()
gocv.CvtColor(digitRegion, &visualization, gocv.ColorGrayToBGR)
for i, box := range digitBoxes {
// Draw box in green
green := color.RGBA{0, 255, 0, 255}
gocv.Rectangle(&visualization, box, green, 2)
// Show box number and size
label := fmt.Sprintf("#%d: %dx%d", i, box.Dx(), box.Dy())
gocv.PutText(&visualization, label, image.Pt(box.Min.X, box.Min.Y-5),
gocv.FontHersheyPlain, 0.8, green, 1)
}
gocv.IMWrite("test_output/debug_digit_boxes.png", visualization)
fmt.Printf("DEBUG: Saved digit boxes to test_output/debug_digit_boxes.png\n")
}
return digitBoxes
}
// ScaleByFactor scales an image by a specific scale factor
func ScaleByFactor(img gocv.Mat, scaleFactor float64) gocv.Mat {
// Calculate new dimensions
newWidth := int(float64(img.Cols()) * scaleFactor)
newHeight := int(float64(img.Rows()) * scaleFactor)
// Resize image
scaled := gocv.NewMat()
gocv.Resize(img, &scaled, image.Point{X: newWidth, Y: newHeight}, 0, 0, gocv.InterpolationLinear)
if DEBUG {
fmt.Printf("DEBUG: Scaled image from %dx%d to %dx%d (scale factor: %.3f)\n",
img.Cols(), img.Rows(), newWidth, newHeight, scaleFactor)
}
return scaled
}
// applyTransforms applies rotation and scaling to a frame
// This is the SINGLE source of truth for transform logic
// Both detection and main loop MUST use this function to ensure identical transforms
func applyTransforms(img gocv.Mat, rotation float64, scaleFactor float64) gocv.Mat {
// Apply rotation (if needed)
var rotated gocv.Mat
if rotation != 0.0 {
rotated = RotateImage(img, rotation)
} else {
rotated = img.Clone()
}
// Apply scaling (if needed)
var scaled gocv.Mat
if scaleFactor != 1.0 {
scaled = ScaleByFactor(rotated, scaleFactor)
if rotation != 0.0 {
rotated.Close() // Only close if we created a new Mat
}
} else {
scaled = rotated
}
return scaled
}
// DisplayAreas holds the two display rectangles for SpO2 and HR
type DisplayAreas struct {
SpO2 image.Rectangle
HR image.Rectangle
}
// MergeDigitBoxesIntoDisplays merges individual digit boxes into two unified display areas
// This is the FINAL step of detection after finding digit boxes
// It splits at the center of the BOUNDING BOX (not image center)
// and merges all left boxes into SpO2, all right boxes into HR display area
func MergeDigitBoxesIntoDisplays(digitRegion gocv.Mat, digitBoxes []image.Rectangle) DisplayAreas {
// First, find the bounding box of ALL digit boxes
contentMinX := digitRegion.Cols()
contentMaxX := 0
for _, box := range digitBoxes {
if box.Min.X < contentMinX {
contentMinX = box.Min.X
}
if box.Max.X > contentMaxX {
contentMaxX = box.Max.X
}
}
// Calculate center of the CONTENT bounding box (not image center)
centerX := (contentMinX + contentMaxX) / 2
if DEBUG {
fmt.Printf("DEBUG: Content bounding box: X[%d-%d], Center X=%d\n",
contentMinX, contentMaxX, centerX)
}
// Initialize bounding boxes
spo2MinX, spo2MaxX := digitRegion.Cols(), 0
spo2MinY, spo2MaxY := digitRegion.Rows(), 0
hrMinX, hrMaxX := digitRegion.Cols(), 0
hrMinY, hrMaxY := digitRegion.Rows(), 0
// Track counts for logging
leftCount := 0
rightCount := 0
// Split boxes by center X and merge
for _, box := range digitBoxes {
// Calculate box center X to determine which half it's in
boxCenterX := (box.Min.X + box.Max.X) / 2
if boxCenterX < centerX {
// LEFT HALF - SpO2
leftCount++
if box.Min.X < spo2MinX {
spo2MinX = box.Min.X
}
if box.Max.X > spo2MaxX {
spo2MaxX = box.Max.X
}
if box.Min.Y < spo2MinY {
spo2MinY = box.Min.Y
}
if box.Max.Y > spo2MaxY {
spo2MaxY = box.Max.Y
}
} else {
// RIGHT HALF - HR
rightCount++
if box.Min.X < hrMinX {
hrMinX = box.Min.X
}
if box.Max.X > hrMaxX {
hrMaxX = box.Max.X
}
if box.Min.Y < hrMinY {
hrMinY = box.Min.Y
}
if box.Max.Y > hrMaxY {
hrMaxY = box.Max.Y
}
}
}
if DEBUG {
fmt.Printf("DEBUG: Split at center X=%d: %d boxes left (SpO2), %d boxes right (HR)\n",
centerX, leftCount, rightCount)
fmt.Printf("DEBUG: SpO2 merged area: X[%d-%d] Y[%d-%d], Size: %dx%d\n",
spo2MinX, spo2MaxX, spo2MinY, spo2MaxY,
spo2MaxX-spo2MinX, spo2MaxY-spo2MinY)
fmt.Printf("DEBUG: HR merged area: X[%d-%d] Y[%d-%d], Size: %dx%d\n",
hrMinX, hrMaxX, hrMinY, hrMaxY,
hrMaxX-hrMinX, hrMaxY-hrMinY)
}
return DisplayAreas{
SpO2: image.Rect(spo2MinX, spo2MinY, spo2MaxX, spo2MaxY),
HR: image.Rect(hrMinX, hrMinY, hrMaxX, hrMaxY),
}
}