File size: 8,651 Bytes
5482ab4
 
 
 
 
 
 
 
 
 
 
46d55ac
 
5482ab4
 
 
 
 
 
 
 
 
 
 
 
 
 
b8d514c
5482ab4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
46d55ac
3fcc88e
46d55ac
 
 
 
3fcc88e
 
 
 
 
 
46d55ac
5482ab4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
419fa8e
 
5482ab4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
419fa8e
5482ab4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c407d5f
5482ab4
 
46d55ac
3fcc88e
 
 
 
5482ab4
 
 
b8d514c
 
 
5482ab4
 
 
b8d514c
5482ab4
 
 
 
b8d514c
 
5482ab4
 
 
 
 
 
 
 
 
 
 
 
 
b8d514c
5482ab4
 
 
 
 
 
b8d514c
5482ab4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b8d514c
5482ab4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
419fa8e
5482ab4
 
 
 
1
2
3
4
5
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
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
import streamlit as st
import cv2
import torch
from PIL import Image
import numpy as np
from transformers import BlipProcessor, BlipForConditionalGeneration
from transformers import ViltProcessor, ViltForQuestionAnswering
import time
from io import BytesIO
import threading
import queue
import os
import tempfile
from datetime import datetime

# Set page config to wide mode
st.set_page_config(layout="wide", page_title="Securade.ai Sentinel")

def initialize_state():
    if 'initialized' not in st.session_state:
        st.session_state.frame = None
        st.session_state.captions = []
        st.session_state.stop_event = threading.Event()
        st.session_state.frame_queue = queue.Queue(maxsize=1)
        st.session_state.caption_queue = queue.Queue(maxsize=10)
        st.session_state.processor = None
        st.session_state.thread = None
        st.session_state.is_streaming = False
        st.session_state.initialized = True

@st.cache_resource
def load_processor():
    class VideoProcessor:
        def __init__(self):
            self.caption_processor = BlipProcessor.from_pretrained("Salesforce/blip-image-captioning-large")
            self.caption_model = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image-captioning-large")
            self.vqa_processor = ViltProcessor.from_pretrained("dandelin/vilt-b32-finetuned-vqa")
            self.vqa_model = ViltForQuestionAnswering.from_pretrained("dandelin/vilt-b32-finetuned-vqa")
            
            # Check for available devices
            if torch.cuda.is_available():
                self.device = "cuda"
            elif torch.backends.mps.is_available():
                self.device = "mps"
            else:
                self.device = "cpu"
            
            self.caption_model.to(self.device)
            self.vqa_model.to(self.device)

        def generate_caption(self, image):
            inputs = self.caption_processor(images=image, return_tensors="pt").to(self.device)
            output = self.caption_model.generate(**inputs, max_new_tokens=50)
            return self.caption_processor.decode(output[0], skip_special_tokens=True)

        def answer_question(self, image, question):
            inputs = self.vqa_processor(image, question, return_tensors="pt").to(self.device)
            outputs = self.vqa_model(**inputs)
            logits = outputs.logits
            idx = logits.argmax(-1).item()
            return self.vqa_model.config.id2label[idx]

    return VideoProcessor()

def get_video_source(source_type, source_path=None):
    if source_type == "Webcam":
        return cv2.VideoCapture(0)
    elif source_type == "Video File" and source_path:
        # Create a temporary file with a specific extension
        temp_dir = tempfile.gettempdir()
        temp_path = os.path.join(temp_dir, 'temp_video.mp4')
        with open(temp_path, 'wb') as f:
            f.write(source_path.getvalue())
        
        cap = cv2.VideoCapture(temp_path)
        if not cap.isOpened():
            st.error("Error: Could not open video file. Please ensure it's a supported format (MP4 with H.264 encoding recommended)")
            return None
        return cap
    elif source_type == "RTSP Stream" and source_path:
        return cv2.VideoCapture(source_path)
    return None

def process_video(stop_event, frame_queue, caption_queue, processor, source_type, source_path=None):
    cap = get_video_source(source_type, source_path)
    last_caption_time = time.time()

    while not stop_event.is_set():
        ret, frame = cap.read()
        if not ret:
            break

        frame = cv2.resize(frame, (800, 600))
        current_time = time.time()

        # Generate caption every 8 seconds
        if current_time - last_caption_time >= 8.0:
            img = Image.fromarray(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB))
            caption = processor.generate_caption(img)
            timestamp = datetime.now().strftime("%H:%M:%S")
            
            try:
                if caption_queue.full():
                    caption_queue.get_nowait()
                caption_queue.put_nowait({'timestamp': timestamp, 'caption': caption})
                last_caption_time = current_time
            except queue.Full:
                pass

        try:
            if frame_queue.full():
                frame_queue.get_nowait()
            frame_queue.put_nowait(frame)
        except queue.Full:
            pass

        # time.sleep(0.03)

    cap.release()

def main():
    initialize_state()
    
    # Main title
    st.title("Securade.ai Sentinel")

    # Create three columns for layout
    video_col, caption_col, qa_col = st.columns([0.4, 0.3, 0.3])

    # Video column
    with video_col:
        st.subheader("Video Feed")
        
        # Video source selection
        source_type = "Video File"
        
        source_path = None
        uploaded_file = None
        if source_type == "Video File":
            uploaded_file = st.file_uploader("Choose a video file", type=['mp4', 'avi', 'mov'])
            if uploaded_file:
                source_path = BytesIO(uploaded_file.getvalue())
        elif source_type == "RTSP Stream":
            source_path = st.text_input("Enter RTSP URL", placeholder="rtsp://your-camera-url")

        start_stop = st.button(
            "Start Surveillance" if not st.session_state.is_streaming else "Stop Surveillance"
        )
        video_placeholder = st.empty()
        
        if start_stop:
            if not st.session_state.is_streaming:
                # Start surveillance
                if st.session_state.processor is None:
                    st.session_state.processor = load_processor()
                st.session_state.stop_event.clear()
                st.session_state.frame_queue = queue.Queue(maxsize=1)
                st.session_state.caption_queue = queue.Queue(maxsize=10)
                st.session_state.thread = threading.Thread(
                    target=process_video,
                    args=(
                        st.session_state.stop_event,
                        st.session_state.frame_queue,
                        st.session_state.caption_queue,
                        st.session_state.processor,
                        source_type,
                        source_path
                    ),
                    daemon=True
                )
                st.session_state.thread.start()
                st.session_state.is_streaming = True
            else:
                # Stop surveillance
                st.session_state.stop_event.set()
                if st.session_state.thread:
                    st.session_state.thread.join(timeout=1.0)
                st.session_state.frame = None
                st.session_state.is_streaming = False
                video_placeholder.empty()

    # Caption column
    with caption_col:
        st.subheader("Scene Analysis")
        caption_placeholder = st.empty()

    # Q&A column
    with qa_col:
        st.subheader("Visual Q&A")
        question = st.text_input("Ask a question about the scene:")
        ask_button = st.button("Ask")
        answer_placeholder = st.empty()

        if ask_button and question and st.session_state.frame is not None:
            img = Image.fromarray(cv2.cvtColor(st.session_state.frame, cv2.COLOR_BGR2RGB))
            answer = st.session_state.processor.answer_question(img, question)
            answer_placeholder.markdown(f"**Answer:** {answer}")

    # Update loop
    if st.session_state.is_streaming:
        placeholder = st.empty()
        while True:
            try:
                # Update video frame
                frame = st.session_state.frame_queue.get_nowait()
                st.session_state.frame = frame
                video_placeholder.image(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB))

                # Update captions
                while not st.session_state.caption_queue.empty():
                    new_caption = st.session_state.caption_queue.get_nowait()
                    st.session_state.captions.append(new_caption)
                    st.session_state.captions = st.session_state.captions[-5:]  # Keep last 5 captions

                if st.session_state.captions:
                    caption_text = "\n\n".join([
                        f"**[{cap['timestamp']}]** {cap['caption']}"
                        for cap in reversed(st.session_state.captions)
                    ])
                    caption_placeholder.markdown(caption_text)

            except queue.Empty:
                # time.sleep(0.01)
                continue

if __name__ == "__main__":
    main()