confedge

A computer vision pipeline built with OpenCV to automatically detect and count cakes placed on an industrial tray along an assembly line. Designed to handle real-time or batch image processing in environments with gray/steel trays and cakes of varying colors and patterns.

⚙️ How It Works

1. Preprocessing

   • Convert input image to grayscale and HSV
   • Apply CLAHE (Contrast Limited Adaptive Histogram Equalization)
   • Gaussian blur to reduce high-frequency noise

2. Tray Segmentation

   • HSV thresholding to detect and mask out steel/black tray regions
   • Inversion to isolate cake candidates

3. Edge Detection + Cleanup

   • Canny edge detection on cleaned mask
   • Morphological dilation and closing to restore object continuity

4. Contour Extraction & Filtering

   • Detect bounding boxes using cv2.findContours
   • Filter based on area, shape, and hierarchy to remove noise

5. Output

   • Count of detected cakes
   • Saved image with bounding boxes and labels (output/annotation.png)

🧪 Sample Usage

Request

You can use curl, Postman, or any HTTP client to send an image.

Example using curl:

curl -X POST "https://productcount.azurewebsites.net/api/count_items" \
  -H "Content-Type: multipart/form-data" \
  -F "image_upload=@sample_images/tray1.jpg"

Requirements

pip install opencv-contrib-python numpy matplotlib azure-functions torch torchvision 

📌 Notes

Assumes the tray is always metallic/gray (HSV-based exclusion)

Designed to work with cakes of any color or pattern

Easily extendable for other industrial object detection use-cases