Hough Circle Detection

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 05.camera_hough_circles.py.

This section downsamples the 640x480 camera feed to 320x240, applies Gaussian blur and Hough circle detection on the grayscale image, then scales the detected circle center and radius coordinates by 2 and draws them back onto the original image.

Principle Explanation

Hough Circle Transform

A circle can be represented by its center (x, y) and radius r. The Hough circle transform votes for possible centers and radii in parameter space based on edge directions in the image; candidates with higher vote counts are returned as circles.

Hough circle computation is intensive, and it can easily combine table edges, reflections, cables, and scattered arcs into false circles. Therefore, this example first downsamples and applies Gaussian blur to reduce pixel count and noisy edges.

Parameter Definitions

circles = cv2.HoughCircles(
    blur, 3,
    dp=1,
    minDist=60,
    param1=100,
    param2=70,
    minRadius=10,
    maxRadius=100
)

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)

Downsample and Preprocess

xxxxxxxxxx
small = cv2.resize(img_np, (320, 240))
gray = cv2.cvtColor(small, cv2.COLOR_BGR2GRAY)
blur = cv2.GaussianBlur(gray, (7, 7), 1.5)

After downsampling, pixel count is about one quarter of the original. The 7x7 Gaussian kernel and sigma=1.5 are used to suppress fragmented edges, but may also weaken very small or blurry circles.

Execute Hough Circle Detection

xxxxxxxxxx
circles = cv2.HoughCircles(
    blur, 3,
    dp=1, minDist=60,
    param1=100, param2=70,
    minRadius=10, maxRadius=100
)

When the return value is None, no circles meeting the conditions were found. The current example reads the first three values as center and radius according to the array layout returned by K230 binding.

Map Coordinates Back and Draw

xxxxxxxxxx
if circles is not None:
    for i in range(circles.shape[0]):
        if i >= 30:
            break
        x = int(circles[i, 0]) * 2
        y = int(circles[i, 1]) * 2
        r = int(circles[i, 2]) * 2
        cv2.circle(img_np, (x, y), 3, (0, 255, 0), -1)
        cv2.circle(img_np, (x, y), r, (0, 0, 255), 2)

Detection is performed at half resolution, so both center and radius are multiplied by 2. The source code draws at most 30 circles to prevent excessive false circles from slowing down display.

Resource Release

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

Parameter Adjustment Notes

• Restrict Radius First: Tightening minRadius and maxRadius according to actual target size is usually more effective than simply raising thresholds.
• Too Many False Detections: Increase param2, minDist, and param1 in order.
• Too Many Missed Detections: Appropriately decrease param2, reduce Gaussian kernel size, confirm radius range covers the target.
• Coordinate Scale: All half-resolution parameters should be understood for 320x240; multiply by 2 when drawing back to 640x480.
• Complex Background: High-reliability applications should add ROI, color masks, contour circularity or circle edge support rate verification; cannot rely solely on a single HoughCircles() call.
• Performance: This algorithm performs global search, so frame rate is usually lower than simple filtering. Disabling IDE image feedback can help distinguish between computation and transmission overhead.