383 lines
11 KiB
Python
383 lines
11 KiB
Python
import argparse
|
|
import re
|
|
import sys
|
|
from pathlib import Path
|
|
|
|
import cv2
|
|
import numpy as np
|
|
import pytesseract
|
|
|
|
|
|
def parse_args():
|
|
ap = argparse.ArgumentParser()
|
|
|
|
ap.add_argument("-v", "--video", default=None,
|
|
help="Percorso video. Se omesso usa webcam 0")
|
|
ap.add_argument("--weights", default="yolov2.weights",
|
|
help="File pesi YOLOv2")
|
|
ap.add_argument("--config", default="yolov2.cfg",
|
|
help="File config YOLOv2")
|
|
ap.add_argument("--labels", default="labels.txt",
|
|
help="File labels classi")
|
|
ap.add_argument("--tesseract-cmd", default=None,
|
|
help="Percorso esplicito a tesseract.exe")
|
|
|
|
# Prestazioni
|
|
ap.add_argument("--input-size", type=int, default=320,
|
|
help="Dimensione input YOLO (320 piu' veloce di 416)")
|
|
ap.add_argument("--detect-every", type=int, default=4,
|
|
help="Esegue YOLO ogni N frame")
|
|
ap.add_argument("--ocr-every-detect", type=int, default=2,
|
|
help="Esegue OCR ogni N cicli di detection")
|
|
ap.add_argument("--preview-width", type=int, default=1280,
|
|
help="Larghezza massima finestra preview")
|
|
ap.add_argument("--max-ocr-boxes", type=int, default=1,
|
|
help="Numero massimo di etichette da leggere per detection")
|
|
|
|
# Qualita'
|
|
ap.add_argument("--min-confidence", type=float, default=0.30,
|
|
help="Soglia minima confidenza")
|
|
ap.add_argument("--label-class", default="etichetta",
|
|
help="Nome classe etichetta")
|
|
ap.add_argument("--min-label-width", type=int, default=50,
|
|
help="Larghezza minima bbox etichetta")
|
|
ap.add_argument("--min-label-height", type=int, default=20,
|
|
help="Altezza minima bbox etichetta")
|
|
ap.add_argument("--ocr-min-digits", type=int, default=2,
|
|
help="Numero minimo di cifre per considerare valida una lettura")
|
|
ap.add_argument("--show-roi", action="store_true",
|
|
help="Mostra ROI preprocessata per OCR")
|
|
ap.add_argument("--print-all", action="store_true",
|
|
help="Stampa anche OCR grezzi non validi")
|
|
|
|
return ap.parse_args()
|
|
|
|
|
|
def require_file(path_str, description):
|
|
path = Path(path_str)
|
|
if not path.exists():
|
|
print(f"Errore: {description} non trovato: {path}")
|
|
sys.exit(1)
|
|
return path
|
|
|
|
|
|
def load_classes(labels_path):
|
|
with open(labels_path, "rt", encoding="utf-8") as f:
|
|
classes = [line.strip() for line in f if line.strip()]
|
|
|
|
if not classes:
|
|
print("Errore: labels.txt vuoto")
|
|
sys.exit(1)
|
|
|
|
return classes
|
|
|
|
|
|
def open_capture(video_arg):
|
|
if video_arg is None:
|
|
cap = cv2.VideoCapture(0, cv2.CAP_DSHOW)
|
|
if not cap.isOpened():
|
|
cap = cv2.VideoCapture(0)
|
|
return cap
|
|
|
|
if str(video_arg).isdigit():
|
|
idx = int(video_arg)
|
|
cap = cv2.VideoCapture(idx, cv2.CAP_DSHOW)
|
|
if not cap.isOpened():
|
|
cap = cv2.VideoCapture(idx)
|
|
return cap
|
|
|
|
return cv2.VideoCapture(video_arg)
|
|
|
|
|
|
def resize_preview(frame, max_width):
|
|
h, w = frame.shape[:2]
|
|
if max_width <= 0 or w <= max_width:
|
|
return frame
|
|
scale = max_width / float(w)
|
|
new_w = int(w * scale)
|
|
new_h = int(h * scale)
|
|
return cv2.resize(frame, (new_w, new_h), interpolation=cv2.INTER_LINEAR)
|
|
|
|
|
|
def clip_box(x1, y1, x2, y2, w, h):
|
|
x1 = max(0, min(x1, w - 1))
|
|
y1 = max(0, min(y1, h - 1))
|
|
x2 = max(0, min(x2, w - 1))
|
|
y2 = max(0, min(y2, h - 1))
|
|
return x1, y1, x2, y2
|
|
|
|
|
|
def expand_box(x1, y1, x2, y2, frame_w, frame_h, pad_ratio=0.08):
|
|
bw = x2 - x1
|
|
bh = y2 - y1
|
|
pad_x = int(bw * pad_ratio)
|
|
pad_y = int(bh * pad_ratio)
|
|
|
|
x1 -= pad_x
|
|
y1 -= pad_y
|
|
x2 += pad_x
|
|
y2 += pad_y
|
|
|
|
return clip_box(x1, y1, x2, y2, frame_w, frame_h)
|
|
|
|
|
|
def preprocess_for_ocr(roi):
|
|
gray = cv2.cvtColor(roi, cv2.COLOR_BGR2GRAY)
|
|
|
|
# Ingrandimento moderato: abbastanza per OCR, meno costoso della versione precedente
|
|
gray = cv2.resize(gray, None, fx=1.7, fy=1.7, interpolation=cv2.INTER_CUBIC)
|
|
|
|
# Leggera pulizia
|
|
gray = cv2.GaussianBlur(gray, (3, 3), 0)
|
|
|
|
# Binaria
|
|
gray = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)[1]
|
|
|
|
# Bordo bianco
|
|
gray = cv2.copyMakeBorder(gray, 8, 8, 8, 8,
|
|
borderType=cv2.BORDER_CONSTANT, value=255)
|
|
|
|
return gray
|
|
|
|
|
|
def ocr_digits_only(roi):
|
|
processed = preprocess_for_ocr(roi)
|
|
|
|
config = r'--oem 3 --psm 7 -c tessedit_char_whitelist=0123456789'
|
|
raw_text = pytesseract.image_to_string(processed, config=config)
|
|
digits = re.sub(r"\D+", "", raw_text)
|
|
|
|
return digits, processed, raw_text
|
|
|
|
|
|
def detect_yolov2_original_style(net, frame, classes, min_confidence):
|
|
"""
|
|
Parsing vicino al tuo script originale.
|
|
Questo evita il parsing 'generico' stile YOLOv3/4 che nel tuo caso
|
|
poteva produrre tantissimi box e rallentare drasticamente tutto.
|
|
"""
|
|
h, w = frame.shape[:2]
|
|
|
|
blob = cv2.dnn.blobFromImage(
|
|
frame,
|
|
scalefactor=1.0 / 255.0,
|
|
size=(args.input_size, args.input_size),
|
|
mean=(0, 0, 0),
|
|
swapRB=False,
|
|
crop=False
|
|
)
|
|
|
|
net.setInput(blob)
|
|
predictions = net.forward()
|
|
|
|
# Adatta forma output
|
|
predictions = np.array(predictions)
|
|
|
|
if predictions.ndim == 4:
|
|
# es. (1, N, 1, 1) o simili: prova a schiacciare
|
|
predictions = predictions.reshape(predictions.shape[1], predictions.shape[-1])
|
|
elif predictions.ndim == 3:
|
|
predictions = predictions[0]
|
|
|
|
detections = []
|
|
|
|
if predictions.ndim != 2:
|
|
return detections
|
|
|
|
for i in range(predictions.shape[0]):
|
|
if predictions.shape[1] <= 5:
|
|
continue
|
|
|
|
prob_arr = predictions[i][5:]
|
|
if prob_arr.size == 0:
|
|
continue
|
|
|
|
class_index = int(prob_arr.argmax(axis=0))
|
|
confidence = float(prob_arr[class_index])
|
|
|
|
if confidence <= min_confidence:
|
|
continue
|
|
|
|
x_center = float(predictions[i][0]) * w
|
|
y_center = float(predictions[i][1]) * h
|
|
width_box = float(predictions[i][2]) * w
|
|
height_box = float(predictions[i][3]) * h
|
|
|
|
x1 = int(x_center - width_box * 0.5)
|
|
y1 = int(y_center - height_box * 0.5)
|
|
x2 = int(x_center + width_box * 0.5)
|
|
y2 = int(y_center + height_box * 0.5)
|
|
|
|
x1, y1, x2, y2 = clip_box(x1, y1, x2, y2, w, h)
|
|
|
|
detections.append({
|
|
"class_id": class_index,
|
|
"label": classes[class_index],
|
|
"confidence": confidence,
|
|
"box": (x1, y1, x2, y2),
|
|
})
|
|
|
|
return detections
|
|
|
|
|
|
def draw_detection(frame, det, extra_text=""):
|
|
x1, y1, x2, y2 = det["box"]
|
|
label = det["label"]
|
|
conf = det["confidence"]
|
|
|
|
color = (255, 255, 255)
|
|
if label.lower() == args.label_class.lower():
|
|
color = (0, 255, 255)
|
|
elif label.lower() == "gaylord":
|
|
color = (0, 255, 0)
|
|
|
|
cv2.rectangle(frame, (x1, y1), (x2, y2), color, 2)
|
|
|
|
text = f"{label} {conf:.2f}"
|
|
if extra_text:
|
|
text += f" | {extra_text}"
|
|
|
|
y_text = max(20, y1 - 8)
|
|
cv2.putText(frame, text, (x1, y_text),
|
|
cv2.FONT_HERSHEY_SIMPLEX, 0.6, color, 2, cv2.LINE_AA)
|
|
|
|
|
|
def best_label_detections(detections, label_name, max_boxes):
|
|
labels = [d for d in detections if d["label"].strip().lower() == label_name.lower()]
|
|
|
|
# Ordina per confidenza e area, cosi' preferisce box piu' leggibili
|
|
labels.sort(
|
|
key=lambda d: (
|
|
d["confidence"],
|
|
(d["box"][2] - d["box"][0]) * (d["box"][3] - d["box"][1])
|
|
),
|
|
reverse=True
|
|
)
|
|
|
|
return labels[:max_boxes]
|
|
|
|
|
|
def main():
|
|
global args
|
|
args = parse_args()
|
|
|
|
if args.tesseract_cmd:
|
|
pytesseract.pytesseract.tesseract_cmd = args.tesseract_cmd
|
|
|
|
require_file(args.weights, "File pesi YOLOv2")
|
|
require_file(args.config, "File config YOLOv2")
|
|
require_file(args.labels, "File labels")
|
|
|
|
classes = load_classes(args.labels)
|
|
print(classes)
|
|
|
|
net = cv2.dnn.readNetFromDarknet(args.config, args.weights)
|
|
net.setPreferableBackend(cv2.dnn.DNN_BACKEND_OPENCV)
|
|
net.setPreferableTarget(cv2.dnn.DNN_TARGET_CPU)
|
|
|
|
cap = open_capture(args.video)
|
|
if not cap.isOpened():
|
|
print("Errore: impossibile aprire la sorgente video")
|
|
sys.exit(1)
|
|
|
|
win_name = "YOLOv2 + OCR etichette (versione alleggerita)"
|
|
cv2.namedWindow(win_name, cv2.WINDOW_NORMAL)
|
|
|
|
frame_idx = 0
|
|
detect_cycle = 0
|
|
|
|
last_detections = []
|
|
last_text_by_box = {}
|
|
printed_texts = {}
|
|
|
|
while True:
|
|
grabbed, frame = cap.read()
|
|
|
|
if not grabbed or frame is None:
|
|
print("Fine stream o impossibile leggere il frame")
|
|
break
|
|
|
|
frame_idx += 1
|
|
|
|
# Detection NON su ogni frame
|
|
if frame_idx == 1 or (frame_idx % args.detect_every == 0):
|
|
detect_cycle += 1
|
|
|
|
last_detections = detect_yolov2_original_style(
|
|
net=net,
|
|
frame=frame,
|
|
classes=classes,
|
|
min_confidence=args.min_confidence
|
|
)
|
|
|
|
label_dets = best_label_detections(
|
|
last_detections,
|
|
args.label_class,
|
|
args.max_ocr_boxes
|
|
)
|
|
|
|
# OCR solo ogni N detection cycles
|
|
if detect_cycle % args.ocr_every_detect == 0:
|
|
new_text_by_box = {}
|
|
|
|
for det in label_dets:
|
|
x1, y1, x2, y2 = det["box"]
|
|
bw = x2 - x1
|
|
bh = y2 - y1
|
|
|
|
if bw < args.min_label_width or bh < args.min_label_height:
|
|
continue
|
|
|
|
rx1, ry1, rx2, ry2 = expand_box(x1, y1, x2, y2, frame.shape[1], frame.shape[0])
|
|
roi = frame[ry1:ry2, rx1:rx2]
|
|
|
|
if roi.size == 0:
|
|
continue
|
|
|
|
digits, processed, raw_text = ocr_digits_only(roi)
|
|
|
|
if args.print_all:
|
|
print(f"[frame {frame_idx}] OCR grezzo='{raw_text.strip()}' -> digits='{digits}'")
|
|
|
|
if len(digits) >= args.ocr_min_digits:
|
|
box_key = (x1, y1, x2, y2)
|
|
new_text_by_box[box_key] = digits
|
|
|
|
last_print_frame = printed_texts.get(digits, -999999)
|
|
if frame_idx - last_print_frame > 30:
|
|
print(f"[frame {frame_idx}] Etichetta letta: {digits}")
|
|
printed_texts[digits] = frame_idx
|
|
|
|
if args.show_roi:
|
|
cv2.imshow("ROI OCR", processed)
|
|
|
|
if new_text_by_box:
|
|
last_text_by_box = new_text_by_box
|
|
|
|
# Disegno su TUTTI i frame, usando l'ultima detection disponibile
|
|
display = frame.copy()
|
|
|
|
for det in last_detections:
|
|
x1, y1, x2, y2 = det["box"]
|
|
box_key = (x1, y1, x2, y2)
|
|
extra = ""
|
|
|
|
if box_key in last_text_by_box:
|
|
extra = f"NUM: {last_text_by_box[box_key]}"
|
|
|
|
draw_detection(display, det, extra_text=extra)
|
|
|
|
display = resize_preview(display, args.preview_width)
|
|
cv2.imshow(win_name, display)
|
|
|
|
key = cv2.waitKey(1) & 0xFF
|
|
if key == ord("q"):
|
|
break
|
|
|
|
cap.release()
|
|
cv2.destroyAllWindows()
|
|
|
|
|
|
if __name__ == "__main__":
|
|
main()
|