Update app.py
Browse files
app.py
CHANGED
@@ -1,51 +1,110 @@
|
|
1 |
-
# app.py
|
2 |
import gradio as gr
|
3 |
import cv2
|
4 |
import numpy as np
|
5 |
-
from
|
|
|
|
|
6 |
|
7 |
-
#
|
8 |
-
|
9 |
-
|
|
|
10 |
|
11 |
-
|
12 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
13 |
|
14 |
-
|
15 |
-
|
16 |
-
|
17 |
-
|
18 |
-
|
|
|
|
|
|
|
|
|
19 |
|
20 |
-
#
|
21 |
-
|
22 |
-
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
29 |
|
30 |
-
#
|
31 |
-
|
32 |
|
33 |
-
#
|
34 |
-
|
35 |
-
|
36 |
-
|
37 |
-
|
38 |
-
|
|
|
|
|
39 |
|
40 |
-
# Gradio interface
|
41 |
iface = gr.Interface(
|
42 |
-
fn=
|
43 |
-
inputs=gr.Video(
|
44 |
-
outputs=gr.Video(),
|
45 |
-
|
46 |
-
|
47 |
)
|
48 |
|
49 |
-
# Launch the
|
50 |
-
|
51 |
-
iface.launch(share=True) # Set share=True to create a public link
|
|
|
|
|
1 |
import gradio as gr
|
2 |
import cv2
|
3 |
import numpy as np
|
4 |
+
from tensorflow.keras.models import load_model
|
5 |
+
from sklearn.preprocessing import StandardScaler
|
6 |
+
from ultralytics import YOLO
|
7 |
|
8 |
+
# Load models
|
9 |
+
lstm_model = load_model('suspicious_activity_model.h5')
|
10 |
+
yolo_model = YOLO('yolov8n-pose.pt') # Ensure this model supports keypoint detection
|
11 |
+
scaler = StandardScaler()
|
12 |
|
13 |
+
def extract_keypoints(frame):
|
14 |
+
"""
|
15 |
+
Extracts normalized keypoints from a frame using YOLO pose model.
|
16 |
+
"""
|
17 |
+
results = yolo_model(frame, verbose=False)
|
18 |
+
for r in results:
|
19 |
+
if r.keypoints is not None and len(r.keypoints) > 0:
|
20 |
+
# Extract the first detected person's keypoints
|
21 |
+
keypoints = r.keypoints.xyn.tolist()[0] # Use the first person's keypoints
|
22 |
+
flattened_keypoints = [kp for keypoint in keypoints for kp in keypoint[:2]] # Flatten x, y values
|
23 |
+
return flattened_keypoints
|
24 |
+
return None # Return None if no keypoints are detected
|
25 |
|
26 |
+
def process_frame(frame):
|
27 |
+
"""
|
28 |
+
Process each frame for suspicious activity detection
|
29 |
+
"""
|
30 |
+
# Perform YOLO detection
|
31 |
+
results = yolo_model(frame, verbose=False)
|
32 |
+
for box in results[0].boxes:
|
33 |
+
cls = int(box.cls[0]) # Class ID
|
34 |
+
confidence = float(box.conf[0])
|
35 |
|
36 |
+
# Detect persons only (class_id 0 for 'person')
|
37 |
+
if cls == 0 and confidence > 0.5:
|
38 |
+
x1, y1, x2, y2 = map(int, box.xyxy[0]) # Bounding box coordinates
|
39 |
+
|
40 |
+
# Extract ROI for classification
|
41 |
+
roi = frame[y1:y2, x1:x2]
|
42 |
+
if roi.size > 0:
|
43 |
+
# Preprocess ROI to extract keypoints
|
44 |
+
keypoints = extract_keypoints(roi)
|
45 |
+
|
46 |
+
if keypoints is not None and len(keypoints) > 0:
|
47 |
+
# Standardize and reshape keypoints for LSTM input
|
48 |
+
keypoints_scaled = scaler.fit_transform([keypoints]) # Standardize features
|
49 |
+
keypoints_reshaped = keypoints_scaled.reshape((1, 1, len(keypoints))) # Reshape for LSTM
|
50 |
+
|
51 |
+
# Predict with LSTM model
|
52 |
+
prediction = (lstm_model.predict(keypoints_reshaped) > 0.5).astype(int)[0][0]
|
53 |
+
|
54 |
+
# Draw bounding box and label
|
55 |
+
color = (0, 0, 255) if prediction == 1 else (0, 255, 0)
|
56 |
+
label = 'Suspicious' if prediction == 1 else 'Normal'
|
57 |
+
cv2.rectangle(frame, (x1, y1), (x2, y2), color, 2)
|
58 |
+
cv2.putText(frame, label, (x1, y1 - 10), cv2.FONT_HERSHEY_SIMPLEX, 0.5, color, 2)
|
59 |
+
else:
|
60 |
+
print("No valid keypoints detected for ROI. Skipping frame.")
|
61 |
+
else:
|
62 |
+
print("ROI size is zero. Skipping frame.")
|
63 |
+
|
64 |
+
return frame
|
65 |
+
|
66 |
+
def detect_suspicious_activity(input_video):
|
67 |
+
"""
|
68 |
+
Main function to process video for suspicious activity detection
|
69 |
+
"""
|
70 |
+
# Open video capture
|
71 |
+
cap = cv2.VideoCapture(input_video)
|
72 |
+
|
73 |
+
# Prepare to save output video
|
74 |
+
width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
|
75 |
+
height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
|
76 |
+
fps = cap.get(cv2.CAP_PROP_FPS)
|
77 |
+
|
78 |
+
# Create VideoWriter object
|
79 |
+
fourcc = cv2.VideoWriter_fourcc(*'mp4v')
|
80 |
+
out = cv2.VideoWriter('output_video.mp4', fourcc, fps, (width, height))
|
81 |
+
|
82 |
+
# Process each frame
|
83 |
+
while cap.isOpened():
|
84 |
+
ret, frame = cap.read()
|
85 |
+
if not ret:
|
86 |
+
break
|
87 |
|
88 |
+
# Process and annotate frame
|
89 |
+
processed_frame = process_frame(frame)
|
90 |
|
91 |
+
# Write processed frame to output video
|
92 |
+
out.write(processed_frame)
|
93 |
+
|
94 |
+
# Release resources
|
95 |
+
cap.release()
|
96 |
+
out.release()
|
97 |
+
|
98 |
+
return 'output_video.mp4'
|
99 |
|
100 |
+
# Create Gradio interface
|
101 |
iface = gr.Interface(
|
102 |
+
fn=detect_suspicious_activity,
|
103 |
+
inputs=gr.Video(label="Upload Video"),
|
104 |
+
outputs=gr.Video(label="Processed Video"),
|
105 |
+
title="Suspicious Activity Detection",
|
106 |
+
description="Upload a video to detect suspicious activities using YOLO and LSTM models"
|
107 |
)
|
108 |
|
109 |
+
# Launch the interface
|
110 |
+
iface.launch()
|
|