tube aligner

This commit is contained in:
2026-04-21 00:19:37 +02:00
parent 42677121e3
commit 04da5da162
24 changed files with 1644 additions and 452 deletions
+315
View File
@@ -0,0 +1,315 @@
'''
Created on 17 avr. 2026
@author: denis
'''
# modules/tube_aligner.py
import cv2
import logging
import numpy as np
logger = logging.getLogger(__name__)
class TubeAligner:
GRBL_THRESHOLD_PX = 20
DEAD_ZONE_PX = 5
def __init__(
self,
px_per_mm : float = 10.0,
grbl_threshold_px : int = 20,
dead_zone_px : int = 5,
debug : bool = False, # ← activable depuis la vue
display = None,
):
self.TUBE_DIAMETER_MM = 16.0
self.grbl_threshold_px = grbl_threshold_px
self.dead_zone_px = dead_zone_px
self.debug = debug
self.display = display
# ------------------------------------------------------------------ #
# Détection principale
# ------------------------------------------------------------------ #
def detect_tube(self, frame: np.ndarray) -> dict:
h, w = frame.shape[:2]
cx_img = w // 2
cy_img = h // 2
result = {
"detected" : False,
"tube_cx" : None,
"tube_cy" : None,
"tube_radius" : None,
"offset_x_px" : 0,
"offset_y_px" : 0,
"offset_x_mm" : 0.0,
"offset_y_mm" : 0.0,
"action" : "none",
"frame_annotated": None,
}
frame_out = frame.copy()
gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
blurred = cv2.GaussianBlur(gray, (15, 15), 3)
# 3 configurations légèrement différentes — vote majoritaire
# Fonctionne sur fond sombre ET fond clair
configs = [
dict(param1=50, param2=30, minRadius=int(min(w,h)*0.26), maxRadius=int(min(w,h)*0.36)),
dict(param1=60, param2=30, minRadius=int(min(w,h)*0.26), maxRadius=int(min(w,h)*0.37)),
dict(param1=50, param2=28, minRadius=int(min(w,h)*0.25), maxRadius=int(min(w,h)*0.365)),
]
all_cx, all_cy, all_r = [], [], []
for cfg in configs:
circles = cv2.HoughCircles(
blurred,
cv2.HOUGH_GRADIENT,
dp=1.2,
minDist=min(w, h) // 2,
**cfg
)
if circles is not None:
c = np.round(circles[0]).astype(int)
best = min(c, key=lambda c: np.sqrt((c[0]-cx_img)**2 + (c[1]-cy_img)**2))
all_cx.append(int(best[0]))
all_cy.append(int(best[1]))
all_r.append(int(best[2]))
if not all_cx:
logger.warning("TubeAligner: aucun cercle détecté (%dx%d)", w, h)
if self.debug:
frame_out = self._draw_debug_no_detection(frame_out, cx_img, cy_img)
result["frame_annotated"] = frame_out
return result
# Moyenne des détections convergentes
tx = int(np.mean(all_cx))
ty = int(np.mean(all_cy))
tr = int(np.mean(all_r))
if tr > 0:
self.px_per_mm = (2 * tr) / 16.0
offset_x_px = tx - cx_img
offset_y_px = ty - cy_img
#offset_x_mm = offset_x_px / self.px_per_mm
#offset_y_mm = offset_y_px /self. px_per_mm
offset_x_mm = offset_y_px /self. px_per_mm # (X CNC = Y image)
offset_y_mm = -offset_x_px / self.px_per_mm # (Y CNC = -X image)
dist_px = float(np.sqrt(offset_x_px**2 + offset_y_px**2))
if dist_px <= self.dead_zone_px:
action = "none"
elif dist_px <= self.grbl_threshold_px:
action = "crop"
else:
action = "grbl"
if self.debug:
frame_out = self._draw_debug(
frame_out, cx_img, cy_img,
tx, ty, tr,
offset_x_px, offset_y_px,
offset_x_mm, offset_y_mm,
dist_px, action,
votes=len(all_cx), # ← affiche le nombre de configs ayant détecté
)
result.update({
"detected" : True,
"tube_cx" : tx,
"tube_cy" : ty,
"tube_radius" : tr,
"offset_x_px" : offset_x_px,
"offset_y_px" : offset_y_px,
"offset_x_mm" : round(offset_x_mm, 3),
"offset_y_mm" : round(offset_y_mm, 3),
"action" : action,
"frame_annotated": frame_out,
})
return result
def crop_to_tube(self, frame: np.ndarray, detection: dict) -> np.ndarray:
"""
Recadrage logiciel : recentre l'image sur le tube détecté.
Utilisé quand action == "crop".
"""
if not detection["detected"]:
return frame
tx = detection["tube_cx"]
ty = detection["tube_cy"]
tr = detection["tube_radius"]
h, w = frame.shape[:2]
# Fenêtre carrée autour du centre du tube
half = tr
x1 = max(tx - half, 0)
y1 = max(ty - half, 0)
x2 = min(tx + half, w)
y2 = min(ty + half, h)
cropped = frame[y1:y2, x1:x2]
# Redimensionne à la taille originale pour ne pas changer le pipeline
return cv2.resize(cropped, (w, h), interpolation=cv2.INTER_LINEAR)
def _detect_center_stable(
self,
capture_func, # callable() → frame bytes
n_samples: int = 5,
delay_s: float = 0.3,
) -> tuple[float, float] | None:
"""
Capture N frames et retourne le centre moyen du tube.
Réduit l'erreur de détection d'un facteur √N.
:param capture_func: callable sans argument → bytes JPEG
:param n_samples: nombre de captures à moyenner
:param delay_s: pause entre chaque capture
:return: (cx_mean, cy_mean) ou None si échec
"""
import time
centers = []
for i in range(n_samples):
if i > 0:
time.sleep(delay_s)
frame_bytes = capture_func()
nparr = np.frombuffer(frame_bytes, np.uint8)
frame = cv2.imdecode(nparr, cv2.IMREAD_COLOR)
if frame is None:
continue
detection = self.detect_tube(frame)
if detection["detected"]:
centers.append((detection["tube_cx"], detection["tube_cy"]))
logger.debug(
"_detect_center_stable [%d/%d] : cx=%d cy=%d",
i+1, n_samples,
detection["tube_cx"], detection["tube_cy"],
)
else:
logger.warning("_detect_center_stable [%d/%d] : tube non détecté", i+1, n_samples)
if len(centers) < 3:
logger.error("_detect_center_stable : seulement %d détections valides", len(centers))
return None
# Filtre les valeurs aberrantes (écart > 2 sigma)
cx_arr = np.array([c[0] for c in centers], dtype=float)
cy_arr = np.array([c[1] for c in centers], dtype=float)
cx_mean, cx_std = np.mean(cx_arr), np.std(cx_arr)
cy_mean, cy_std = np.mean(cy_arr), np.std(cy_arr)
mask = (
(np.abs(cx_arr - cx_mean) <= 2 * cx_std) &
(np.abs(cy_arr - cy_mean) <= 2 * cy_std)
)
filtered = [(cx_arr[i], cy_arr[i]) for i in range(len(centers)) if mask[i]]
if not filtered:
filtered = centers # fallback si tout est filtré
cx_final = float(np.mean([c[0] for c in filtered]))
cy_final = float(np.mean([c[1] for c in filtered]))
logger.info(
"_detect_center_stable : %d/%d valides cx=%.1f±%.1f cy=%.1f±%.1f",
len(filtered), n_samples,
cx_final, cx_std, cy_final, cy_std,
)
return cx_final, cy_final
def calib_reset(self):
pass
# ------------------------------------------------------------------ #
# Dessin debug
# ------------------------------------------------------------------ #
def _draw_debug(
self, frame, cx_img, cy_img,
tx, ty, tr,
offset_x_px, offset_y_px,
offset_x_mm, offset_y_mm,
dist_px, action, votes: int = 3
) -> np.ndarray:
# Couleur selon action
color = {
"none" : (0, 255, 0), # vert — centré
"crop" : (0, 200, 255), # orange — recadrage
"grbl" : (0, 0, 255), # rouge — correction CNC
}.get(action, (200, 200, 200))
# Cercle intérieur du tube
cv2.circle(frame, (tx, ty), tr, color, 2, cv2.LINE_AA)
# Rayon de zone morte (dead zone) en vert clair
cv2.circle(frame, (cx_img, cy_img), self.dead_zone_px,
(0, 255, 100), 1, cv2.LINE_AA)
# Rayon seuil GRBL en rouge pointillé (simulé par cercle fin)
cv2.circle(frame, (cx_img, cy_img), self.grbl_threshold_px,
(0, 80, 255), 1, cv2.LINE_AA)
# Croix centre image
cv2.drawMarker(frame, (cx_img, cy_img),
(255, 255, 255), cv2.MARKER_CROSS, 24, 1, cv2.LINE_AA)
# Centre tube
cv2.circle(frame, (tx, ty), 5, color, -1, cv2.LINE_AA)
# Vecteur offset centre image → centre tube
if dist_px > self.dead_zone_px:
cv2.arrowedLine(frame, (cx_img, cy_img), (tx, ty),
color, 2, cv2.LINE_AA, tipLength=0.2)
# Panneau info — fond semi-transparent
overlay = frame.copy()
cv2.rectangle(overlay, (8, 8), (400, 130), (0, 0, 0), -1)
cv2.addWeighted(overlay, 0.45, frame, 0.55, 0, frame)
lines = [
(f"Tube cx={tx} cy={ty} r={tr}px", (0, 255, 180)),
(f"Offset dx={offset_x_px:+d}px dy={offset_y_px:+d}px", color),
(f"Offset dx={offset_x_mm:+.3f}mm dy={offset_y_mm:+.3f}mm", color),
(f"Dist={dist_px:.1f}px action={action.upper()}", color),
(f"px/mm={self.px_per_mm:.4f} votes={votes}/3", (180, 180, 180)), # ← votes
]
for i, (text, col) in enumerate(lines):
cv2.putText(frame, text, (14, 30 + i * 20),
cv2.FONT_HERSHEY_SIMPLEX, 0.48, col, 1, cv2.LINE_AA)
# Légende zones
cv2.putText(frame, "dead zone", (cx_img + self.dead_zone_px + 3, cy_img - 3),
cv2.FONT_HERSHEY_SIMPLEX, 0.35, (0, 255, 100), 1)
cv2.putText(frame, "GRBL threshold", (cx_img + self.grbl_threshold_px + 3, cy_img + 6),
cv2.FONT_HERSHEY_SIMPLEX, 0.35, (0, 80, 255), 1)
return frame
def _draw_debug_no_detection(self, frame, cx_img, cy_img) -> np.ndarray:
cv2.drawMarker(frame, (cx_img, cy_img),
(255, 255, 255), cv2.MARKER_CROSS, 24, 1, cv2.LINE_AA)
cv2.putText(frame, "Tube non detecte", (14, 30),
cv2.FONT_HERSHEY_SIMPLEX, 0.6, (0, 0, 255), 2, cv2.LINE_AA)
return frame