TheKnight115's picture
Update app.py
f0a70b1 verified
raw
history blame
14.8 kB
import streamlit as st
import cv2
import numpy as np
import tempfile
import time
from ultralytics import YOLO
from huggingface_hub import hf_hub_download
from email.mime.text import MIMEText
from email.mime.multipart import MIMEMultipart
from email.mime.base import MIMEBase
from email import encoders
import os
import smtplib
from transformers import AutoModel, AutoProcessor
from PIL import Image, ImageDraw, ImageFont
import re
import torch
# Email credentials
FROM_EMAIL = "[email protected]"
EMAIL_PASSWORD = "cawxqifzqiwjufde" # App-Specific Password
TO_EMAIL = "[email protected]"
SMTP_SERVER = 'smtp.gmail.com'
SMTP_PORT = 465
# Arabic dictionary for converting license plate text
arabic_dict = {
"0": "٠", "1": "١", "2": "٢", "3": "٣", "4": "٤", "5": "٥",
"6": "٦", "7": "٧", "8": "٨", "9": "٩", "A": "ا", "B": "ب",
"J": "ح", "D": "د", "R": "ر", "S": "س", "X": "ص", "T": "ط",
"E": "ع", "G": "ق", "K": "ك", "L": "ل", "Z": "م", "N": "ن",
"H": "ه", "U": "و", "V": "ي", " ": " "
}
# Color mapping for different classes
class_colors = {
0: (0, 255, 0), # Green (Helmet)
1: (255, 0, 0), # Blue (License Plate)
2: (0, 0, 255), # Red (MotorbikeDelivery)
3: (255, 255, 0), # Cyan (MotorbikeSport)
4: (255, 0, 255), # Magenta (No Helmet)
5: (0, 255, 255), # Yellow (Person)
}
# Load the OCR model
processor = AutoProcessor.from_pretrained("stepfun-ai/GOT-OCR2_0", trust_remote_code=True)
model_ocr = AutoModel.from_pretrained("stepfun-ai/GOT-OCR2_0", trust_remote_code=True).to('cuda')
# Define lane area coordinates (example coordinates)
red_lane = np.array([[2, 1583], [1, 1131], [1828, 1141], [1912, 1580]], np.int32)
# YOLO inference function
def run_yolo(image):
results = model(image)
return results
# Function to process YOLO results and draw bounding boxes
def process_results(results, image):
boxes = results[0].boxes
for box in boxes:
x1, y1, x2, y2 = map(int, box.xyxy[0])
conf = box.conf[0]
cls = int(box.cls[0])
label = model.names[cls]
color = class_colors.get(cls, (255, 255, 255))
# Draw rectangle and label
cv2.rectangle(image, (x1, y1), (x2, y2), color, 2)
cv2.putText(image, f"{label} {conf:.2f}", (x1, y1 - 10), cv2.FONT_HERSHEY_SIMPLEX, 0.5, color, 2)
return image
# Process uploaded images
def process_image(uploaded_file):
image = np.array(cv2.imdecode(np.frombuffer(uploaded_file.read(), np.uint8), 1))
results = run_yolo(image)
processed_image = process_results(results, image)
processed_image_rgb = cv2.cvtColor(processed_image, cv2.COLOR_BGR2RGB)
st.image(processed_image_rgb, caption='Detected Image', use_column_width=True)
# Create a download button for the processed image
im_pil = Image.fromarray(processed_image_rgb)
im_pil.save("processed_image.png")
with open("processed_image.png", "rb") as file:
btn = st.download_button(
label="Download Processed Image",
data=file,
file_name="processed_image.png",
mime="image/png"
)
# Process and save uploaded videos
@st.cache_data
# Define the function to process the video
def process_video_and_save(uploaded_file):
# Path for Arabic font
font_path = "alfont_com_arial-1.ttf"
# Paths for saving violation images
violation_image_path = 'violation.jpg'
# Track emails already sent to avoid duplicate emails
sent_emails = {}
# Dictionary to track violations per license plate
violations_dict = {}
# Paths for saving violation images and videos
video_path = "uploaded_video.mp4"
output_video_path = 'output_violation.mp4'
# Save the uploaded video file to this path
with open(video_path, "wb") as f:
f.write(uploaded_file.getbuffer())
cap = cv2.VideoCapture(video_path)
if not cap.isOpened():
st.error("Error opening video file.")
return None
# Codec and output settings
fourcc = cv2.VideoWriter_fourcc(*'mp4v')
fps = cap.get(cv2.CAP_PROP_FPS)
width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
out = cv2.VideoWriter(output_video_path, fourcc, fps, (width, height))
margin_y = 50
# Process frames
while cap.isOpened():
ret, frame = cap.read()
if not ret:
break # End of video
# Draw the red lane rectangle on each frame
cv2.polylines(frame, [red_lane], isClosed=True, color=(0, 0, 255), thickness=3) # Red lane
# Perform detection using YOLO on the current frame
results = model.track(frame)
# Process each detection in the results
for box in results[0].boxes:
x1, y1, x2, y2 = map(int, box.xyxy[0].cpu().numpy()) # Bounding box coordinates
label = model.names[int(box.cls)] # Class name (MotorbikeDelivery, Helmet, etc.)
color = (255, 0, 0) # Use a fixed color for bounding boxes
confidence = box.conf[0].item()
# Initialize flags and variables for the violations
helmet_violation = False
lane_violation = False
violation_type = []
# Draw bounding box around detected object
cv2.rectangle(frame, (x1, y1), (x2, y2), color, 3) # 3 is the thickness of the rectangle
# Add label to the box (e.g., 'MotorbikeDelivery')
cv2.putText(frame, f'{label}: {confidence:.2f}', (x1, y1 - 10), cv2.FONT_HERSHEY_SIMPLEX, 0.6, color, 2)
# Detect MotorbikeDelivery
if label == 'MotorbikeDelivery' and confidence >= 0.4:
motorbike_crop = frame[max(0, y1 - margin_y):y2, x1:x2]
delivery_center = ((x1 + x2) // 2, (y2))
in_red_lane = cv2.pointPolygonTest(red_lane, delivery_center, False)
if in_red_lane >= 0:
lane_violation = True
violation_type.append("In Red Lane")
# Perform detection within the cropped motorbike region
sub_results = model(motorbike_crop)
for result in sub_results[0].boxes:
sub_x1, sub_y1, sub_x2, sub_y2 = map(int, result.xyxy[0].cpu().numpy()) # Bounding box coordinates
sub_label = model.names[int(result.cls)]
sub_color = (255, 0, 0) # Red color for the bounding box of sub-objects
# Draw bounding box around sub-detected objects (No_Helmet, License_plate, etc.)
cv2.rectangle(motorbike_crop, (sub_x1, sub_y1), (sub_x2, sub_y2), sub_color, 2)
cv2.putText(motorbike_crop, sub_label, (sub_x1, sub_y1 - 10), cv2.FONT_HERSHEY_SIMPLEX, 0.6, sub_color, 2)
if sub_label == 'No_Helmet':
helmet_violation = True
violation_type.append("No Helmet")
continue
if sub_label == 'License_plate':
license_crop = motorbike_crop[sub_y1:sub_y2, sub_x1:sub_x2]
# Apply OCR if a violation is detected
if helmet_violation or lane_violation:
# Perform OCR on the license plate
cv2.imwrite(violation_image_path, frame)
license_plate_pil = Image.fromarray(cv2.cvtColor(license_crop, cv2.COLOR_BGR2RGB))
temp_image_path = 'license_plate.png'
license_plate_pil.save(temp_image_path)
license_plate_text = model_ocr.chat(processor, temp_image_path, ocr_type='ocr')
filtered_text = filter_license_plate_text(license_plate_text)
# Check if the license plate is already detected and saved
if filtered_text:
# Add the license plate and its violations to the violations dictionary
if filtered_text not in violations_dict:
violations_dict[filtered_text] = violation_type #{"1234AB":[no_Helmet,In_red_Lane]}
send_email(filtered_text, violation_image_path, ', '.join(violation_type))
else:
# Update the violations for the license plate if new ones are found
current_violations = set(violations_dict[filtered_text]) # no helmet
new_violations = set(violation_type) # red lane, no helmet
updated_violations = list(current_violations | new_violations) # red_lane, no helmet
# If new violations are found, update and send email
if updated_violations != violations_dict[filtered_text]:
violations_dict[filtered_text] = updated_violations
send_email(filtered_text, violation_image_path, ', '.join(updated_violations))
# Draw OCR text (English and Arabic) on the original frame
arabic_text = convert_to_arabic(filtered_text)
frame = draw_text_pil(frame, filtered_text, (x1, y2 + 30), font_path, font_size=30, color=(255, 255, 255))
frame = draw_text_pil(frame, arabic_text, (x1, y2 + 60), font_path, font_size=30, color=(0, 255, 0))
# Write the processed frame to the output video
out.write(frame)
# Release resources when done
cap.release()
out.release()
if not os.path.exists(output_video_path):
st.error("Error: Processed video was not created.")
return output_video_path # Return the path of the processed video
# Live video feed processing
def live_video_feed():
stframe = st.empty()
video = cv2.VideoCapture(0)
if not video.isOpened():
st.error("Unable to access the webcam.")
return
while True:
ret, frame = video.read()
if not ret:
st.error("Failed to capture frame.")
break
# Run YOLO on the captured frame
results = run_yolo(frame)
annotated_frame = process_results(results, frame)
annotated_frame_rgb = cv2.cvtColor(annotated_frame, cv2.COLOR_BGR2RGB)
# Display the frame with detections
stframe.image(annotated_frame_rgb, channels="RGB", use_column_width=True)
if st.button("Stop"):
break
video.release()
st.stop()
# Function to filter license plate text
def filter_license_plate_text(license_plate_text):
license_plate_text = re.sub(r'[^A-Z0-9]+', "", license_plate_text)
match = re.search(r'(\d{3,4})\s*([A-Z]{2})', license_plate_text)
return f"{match.group(1)} {match.group(2)}" if match else None
# Function to convert license plate text to Arabic
def convert_to_arabic(license_plate_text):
return "".join(arabic_dict.get(char, char) for char in license_plate_text)
# Function to send email notification with image attachment
def send_email(license_text, violation_image_path, violation_type):
if violation_type == 'no_helmet':
subject = 'تنبيه مخالفة: عدم ارتداء خوذة'
body = f"لعدم ارتداء الخوذة ({license_text}) تم تغريم دراجة نارية التي تحمل لوحة."
elif violation_type == 'in_red_lane':
subject = 'تنبيه مخالفة: دخول المسار الأيسر'
body = f"لدخولها المسار الأيسر ({license_text}) تم تغريم دراجة نارية التي تحمل لوحة."
elif violation_type == 'no_helmet_in_red_lane':
subject = 'تنبيه مخالفة: عدم ارتداء خوذة ودخول المسار الأيسر'
body = f"لعدم ارتداء الخوذة ولدخولها المسار الأيسر ({license_text}) تم تغريم دراجة نارية التي تحمل لوحة."
msg = MIMEMultipart()
msg['From'] = FROM_EMAIL
msg['To'] = TO_EMAIL
msg['Subject'] = subject
msg.attach(MIMEText(body, 'plain'))
if os.path.exists(violation_image_path):
with open(violation_image_path, 'rb') as attachment_file:
part = MIMEBase('application', 'octet-stream')
part.set_payload(attachment_file.read())
encoders.encode_base64(part)
part.add_header('Content-Disposition', f'attachment; filename={os.path.basename(violation_image_path)}')
msg.attach(part)
with smtplib.SMTP_SSL(SMTP_SERVER, SMTP_PORT) as server:
server.login(FROM_EMAIL, EMAIL_PASSWORD)
server.sendmail(FROM_EMAIL, TO_EMAIL, msg.as_string())
print("Email with attachment sent successfully!")
def draw_text_pil(img, text, position, font_path, font_size, color):
img_pil = Image.fromarray(cv2.cvtColor(img, cv2.COLOR_BGR2RGB))
draw = ImageDraw.Draw(img_pil)
try:
font = ImageFont.truetype(font_path, size=font_size)
except IOError:
print(f"Font file not found at {font_path}. Using default font.")
font = ImageFont.load_default()
draw.text(position, text, font=font, fill=color)
img_np = cv2.cvtColor(np.array(img_pil), cv2.COLOR_RGB2BGR)
return img_np
# Streamlit app main function
def main():
model_file = hf_hub_download(repo_id="TheKnight115/Yolov8m", filename="yolov8_Medium.pt")
global model
model = YOLO(model_file)
st.title("Motorbike Violation Detection")
input_type = st.selectbox("Select Input Type", ("Image", "Video", "Live Feed"))
if input_type == "Image":
uploaded_file = st.file_uploader("Choose an image...", type=["jpg", "jpeg", "png"])
if uploaded_file is not None:
process_image(uploaded_file)
elif input_type == "Video":
uploaded_file = st.file_uploader("Choose a video...", type=["mp4", "mov"])
if uploaded_file is not None:
output_path = process_video_and_save(uploaded_file)
# Now, move the download button here, outside the cached function
with open(output_path, "rb") as video_file:
btn = st.download_button(
label="Download Processed Video",
data=video_file,
file_name="processed_video.mp4",
mime="video/mp4"
)
elif input_type == "Live Feed":
live_video_feed()
if __name__ == "__main__":
main()