Machine Vision

// HAMK Machine Vision Course · Python + OpenCV + RoboDK + Dobot MG400 · Group 11

Task A — RGB Channel Separation (OpenCV)

Original Image (BGR→RGB)

Red Channel

Green Channel

Blue Channel

        Image shape: 480 × 640 × 3 · 
        dtype: uint8 · 
        Channel order: BGR (OpenCV) → RGB (matplotlib)
      

Pixel Intensity Histogram

Segmentation Pipeline — Step by Step

① Original

② Grayscale

③ CLAHE+Blur

④ Otsu Threshold

⑤ Morphology

⑥ Detected Objects

Detected Objects — connectedComponentsWithStats

Histogram (Grayscale)

Otsu threshold: — · Objects kept: — · Min area: 1244 px

Camera → Robot Coordinate Mapping (Homography)

Click anywhere on the image to map pixel coordinates → Dobot MG400 robot coordinates. 8-point calibration, cv2.findHomography().

📌 Click to get robot coordinates · 8 calibration tiles shown

Last Click

Pixel: —
Robot X: — mm
Robot Y: — mm

Calibration Data (8 pts)

ID	Robot X	Robot Y	px u	px v

Homography Matrix H

Mean error: 0.8 mm · Max error: 1.9 mm

Segmentation Pipeline — Python + OpenCV

import cv2
import numpy as np
import matplotlib.pyplot as plt

# ── 1. Load and convert ──────────────────────────────
img     = cv2.imread('robodk_snapshot.png')
img_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
gray    = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

# ── 2. Contrast enhancement (CLAHE) + denoise ────────
clahe      = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8,8))
gray_clahe = clahe.apply(gray)
blurred    = cv2.GaussianBlur(gray_clahe, (5,5), 0)

# ── 3. Otsu thresholding (objects darker than BG) ────
_, binary = cv2.threshold(
    blurred, 0, 255,
    cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU
)

# ── 4. Morphological cleanup ─────────────────────────
kernel  = cv2.getStructuringElement(cv2.MORPH_RECT, (5,5))
opened  = cv2.morphologyEx(binary, cv2.MORPH_OPEN,  kernel)
cleaned = cv2.morphologyEx(opened, cv2.MORPH_CLOSE, kernel)

# ── 5. Connected components ──────────────────────────
num_labels, labeled, stats, centroids = cv2.connectedComponentsWithStats(
    cleaned, connectivity=4, ltype=cv2.CV_32S
)
MIN_AREA = 1244   # px — area filter threshold
kept = [(stats[i], centroids[i]) for i in range(1, num_labels)
        if stats[i][cv2.CC_STAT_AREA] >= MIN_AREA]
print(f"Found {len(kept)} objects")

# ── 6. Annotate ──────────────────────────────────────
annotated = img_rgb.copy()
for (stat, (cx, cy)) in kept:
    x, y, w, h = stat[:4]
    cv2.rectangle(annotated, (x,y), (x+w,y+h), (0,255,0), 2)
    cv2.circle(annotated, (int(cx),int(cy)), 5, (255,0,0), -1)
    cv2.putText(annotated, f"({int(cx)},{int(cy)})",
               (x+2, y-6), cv2.FONT_HERSHEY_SIMPLEX,
               0.4, (0,255,0), 1)

Homography — pixel_to_robot(u, v, H)

import numpy as np
import cv2

# Calibration points (8 tiles at known robot positions)
img_pts   = np.array([[604,717],[609,1235],[997,951],[721,468],
                       [476,482],[484,973],[847,586],[740,962]], dtype=np.float32)
robot_pts = np.array([[350,0],[350,100],[275,50],[325,-50],
                       [375,-75],[375,50],[300,-25],[325,50]], dtype=np.float32)

H, _ = cv2.findHomography(img_pts, robot_pts, method=0)

def pixel_to_robot(u, v, H):
    p  = np.array([u, v, 1.0], dtype=np.float32).reshape(3,1)
    pr = H @ p
    pr = pr / pr[2,0]      # homogeneous → Euclidean
    return pr[0,0], pr[1,0]  # (X_mm, Y_mm)

# Mean calibration error: 0.8 mm — Max: 1.9 mm