Background

I have a goal of stacking Jenga blocks. But before the robot arm can do anything fancy, I need to stop being scared of images.

Short Goal

So I started with this static photo of Jenga blocks from my desktop camera. The small mission: draw accurate rotated boxes, center dots, and labels around the blocks by the end of this post.

Original RealSense camera image with five Jenga blocks on a table — The starting photo: five blocks, 640 x 480 pixels, shot with an Intel RealSense D435. The lighting is not perfect, the shadow is dramatic, and OpenCV is about to judge me.

After some trial, error, and mild emotional damage, this became the first usable output:

Detected Jenga blocks with green rotated boxes, center dots, and angle labels — Green boxes, center dots, sizes, and angles. The blocks are mostly found. The shadow on the right is still trying to become famous.

And the stupidity begins…

OpenCV 1

1 2	import cv2 as lv import numpy as np

Typically, people import cv2 as cv, but why not make it un poco luxurious? lv sounds richer to me. HAHAH!

Do Not Be Afraid by OpenClaw

OpenClaw once said, “yeah, I got you.”

📄 opencv.pdf - OpenClaw's "yeah, I got you" notes

Reference Sample

Check this only if you’re out of time. For me, I find it useless for personal growth and critical thinking.

Show/hide sample Python code

import cv2
import numpy as np

# === LOAD IMAGE ===
img = cv2.imread('cv.png')
h, w = img.shape[:2]
print(f"Image: {w}x{h}")

# === STEP 1: HSV COLOR SEGMENTATION ===
# Wooden blocks: warm hue (H < 20), moderate-to-high saturation (S > 25)
# Saturation is the key — table is desaturated, blocks have color
hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
lower = np.array([0, 25, 0])   # H, S, V
upper = np.array([20, 255, 255])
mask = cv2.inRange(hsv, lower, upper)

# === STEP 2: MORPHOLOGY — clean up mask ===
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (5, 5))
mask = cv2.morphologyEx(mask, cv2.MORPH_CLOSE, kernel, iterations=2)
mask = cv2.morphologyEx(mask, cv2.MORPH_OPEN, kernel, iterations=1)

# === STEP 3: GAUSSIAN BLUR before Canny ===
# Smooths noise and wood grain → cleaner edge detection
blurred = cv2.GaussianBlur(mask, (5, 5), 0)

# === STEP 4: CANNY EDGE DETECTION ===
edges = cv2.Canny(blurred, 50, 150)
edges = cv2.morphologyEx(edges, cv2.MORPH_CLOSE, kernel, iterations=2)

# === STEP 5: FIND CONTOURS ===
contours, _ = cv2.findContours(edges, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)

result = img.copy()
blocks = []

for cnt in contours:
    area = cv2.contourArea(cnt)
    if area < 1000:
        continue  # noise threshold

    # === STEP 6: ROTATED BOUNDING BOX ===
    # minAreaRect returns: (center=(cx,cy), size=(w,h), angle=degrees)
    # angle is relative to the horizontal axis
    rect = cv2.minAreaRect(cnt)
    box = cv2.boxPoints(rect).astype(int)
    rw, rh = rect[1]
    aspect = rw / float(rh) if rh > 0 else 0

    # Filter by size/aspect ratio
    if 0.2 < aspect < 6.0 and area > 2000:
        cx, cy = int(rect[0][0]), int(rect[0][1])
        blocks.append({"rect": rect, "area": area})

        # Draw rotated rectangle
        cv2.drawContours(result, [box], 0, (0, 255, 0), 2)
        # Center dot
        cv2.circle(result, (cx, cy), 4, (0, 255, 0), -1)
        # Label: width x height @ angle
        label = f"{rw:.0f}x{rh:.0f}  @{rect[2]:.1f}deg"
        cv2.putText(result, label, (cx - 45, cy - 12),
                    cv2.FONT_HERSHEY_SIMPLEX, 0.35, (0, 255, 0), 1)

print(f"\nPipeline: HSV → GaussianBlur(5x5) → Canny(50,150) → minAreaRect")
print(f"Found {len(blocks)} blocks:")
for i, b in enumerate(blocks):
    r = b["rect"]
    print(f"  Block {i+1}: {r[1][0]:.0f}x{r[1][1]:.0f} "
          f"at ({r[0][0]:.0f},{r[0][1]:.0f}), "
          f"angle={r[2]:.1f}deg, area={b['area']:.0f}")

# === SAVE ===
# cv2.imwrite('cv_detected.png', result)
# cv2.imwrite('cv_edges.png', edges)

cv2.imshow('result', result)
cv2.waitKey(0)        # wait for a key press
cv2.destroyAllWindows()  # close the window