Contour Analysis

Preface

This tutorial is applicable to firmware version: CanMV_K230_YAHBOOM_micropython_v1.8-13. Download from Jianan official website https://download.kendryte.com/developer/releases/canmv_k230_micropython/daily_build/

Example Results

Run the example code for this section 07.camera_find_contours.py.

This section performs grayscale conversion, Gaussian denoising, fixed threshold binarization, and closing operation on a 320x240 image, extracts external contours, calculates area, perimeter, polygon approximation and bounding rectangle, then scales the results by 2 and draws them onto the 640x480 original image.

Principle Explanation

What are Contours

Contours are a set of pixel coordinates arranged in boundary order. In binary images, white regions are typically treated as foreground, and findContours() generates contour point sets along the foreground boundary. Contours can further calculate area, perimeter, bounding rectangle, and approximating polygon.

This example's processing flow is as follows:

640x480 BGR image
      ↓ resize
320x240 BGR image
      ↓ grayscale, Gaussian blur, fixed threshold
Binary image
      ↓ closing operation
Foreground with holes filled
      ↓ findContours
External contours
      ↓ area, perimeter, approximation, bounding rectangle
      ↓ coordinates multiplied by 2
Draw onto original image

Main Contour Properties

• contourArea(): Calculates the area enclosed by the contour.

• arcLength(): Calculates the contour perimeter, the second parameter indicates whether it is closed.

• approxPolyDP(): Simplifies the contour according to the error threshold.

• boundingRect(): Returns the axis-aligned bounding rectangle (x, y, w, h).

Code Overview

Initialize Camera and Display Module

xxxxxxxxxx
W, H = 640, 480
sensor = Sensor(width=1280, height=960, fps=90)
sensor.reset()
sensor.set_framesize(width=W, height=H)
sensor.set_pixformat(Sensor.RGB888)
Display.init(Display.ST7701, width=640, height=480, to_ide=True)
sensor.run()
time.sleep(0.5)

Set Area Threshold and Morphological Kernel

xxxxxxxxxx
MIN_A = 50
k = cv2.getStructuringElement(cv2.MORPH_RECT, (3, 3))

MIN_A=50 is the area threshold set for 320x240 processed image. When scaling back to 640x480 full resolution processing, the area scale expands to approximately four times, so this value cannot be used directly.

Image Preprocessing

xxxxxxxxxx
small = cv2.resize(img_np, (320, 240))
gray = cv2.cvtColor(small, cv2.COLOR_BGR2GRAY)
blur = cv2.GaussianBlur(gray, (3, 3), 0)
ret, binary = cv2.threshold(blur, 127, 255, cv2.THRESH_BINARY)
closed = cv2.morphologyEx(binary, cv2.MORPH_CLOSE, k)

The source code actually uses fixed threshold 127, not OTSU. Closing operation is used to fill small holes and boundary gaps inside the target.

Find and Analyze Contours

​x
contours, _ = cv2.findContours(
    closed,
    cv2.RETR_EXTERNAL,
    cv2.CHAIN_APPROX_SIMPLE
)

for contour in contours:
    area = cv2.contourArea(contour)
    if area < MIN_A:
        continue

    perimeter = cv2.arcLength(contour, True)
    approx = cv2.approxPolyDP(contour, 0.02 * perimeter, True)
    x, y, w, h = cv2.boundingRect(contour)

Only external contours are extracted, with a maximum of 50 valid contours processed. Approximation error is 2% of the perimeter.

Map Coordinates Back and Draw

xxxxxxxxxx
cv2.drawContours(img_np, [approx * 2], -1, (0, 255, 0), 2)
cv2.rectangle(
    img_np,
    (x * 2, y * 2),
    (x * 2 + w * 2, y * 2 + h * 2),
    (0, 0, 255), 1
)

Point coordinates, rectangle origin, width and height all need to be multiplied by 2. If only scaling the origin but forgetting to scale the width and height, the drawn rectangle will be noticeably smaller.

Resource Release

xxxxxxxxxx
finally:
    sensor.stop()
    Display.deinit()
    os.exitpoint(os.EXITPOINT_ENABLE_SLEEP)
    time.sleep_ms(100)

Parameter Adjustment Notes

• Fixed Threshold 127: Needs to be reset when the target is too dark or lighting varies significantly. Can also switch to OTSU and re-verify.

• MIN_A: Currently calculated based on 320x240 area; should be adjusted proportionally by pixel area when processing resolution changes.

• Closing Kernel: Can be changed to 5x5 when gaps are large, but larger kernels are more likely to connect adjacent targets.

• Approximation Error 0.02: Increasing reduces vertices, decreasing preserves more boundary details.

• External Contour Mode: When analyzing holes or nested contours, RETR_EXTERNAL is insufficient. Choose a retrieval mode that supports hierarchy and process hierarchy.

• Coordinate Mapping: If subsequent processing size is not fixed at half, calculate scaling ratios separately using W/processed_width and H/processed_height.