backend/dashboard.py

"""
HaramGuard — Dashboard Entry Point
=====================================
"""

import cv2
import streamlit as st
import numpy as np

from config   import VIDEO_PATH, DB_PATH, MODEL_PATH, OPENAI_KEY, ANTHROPIC_KEY
from pipeline import RealTimePipeline

# ── Init ──────────────────────────────────────────────────────────────
st.set_page_config(page_title='HaramGuard', layout='wide')
st.title('🕌 HaramGuard — Real-Time Crowd Safety')

pipeline = RealTimePipeline(
    video_path    = VIDEO_PATH,
    openai_key    = OPENAI_KEY,
    anthropic_key = ANTHROPIC_KEY,   # enables Claude Vision counting
    model_path    = MODEL_PATH,
    db_path       = DB_PATH,
)
cap = pipeline.get_video_capture()

# ── Placeholders ──────────────────────────────────────────────────────
col1, col2 = st.columns([2, 1])

with col1:
    ph_video = st.empty()

with col2:
    ph_risk    = st.empty()
    ph_count   = st.empty()
    ph_trend   = st.empty()
    ph_alert   = st.empty()
    ph_reflect = st.empty()

# ── Main loop ─────────────────────────────────────────────────────────
while cap.isOpened():
    ret, frame = cap.read()
    if not ret:
        break

    pipeline.process_one_frame(frame)
    s = pipeline.state

    # Video feed
    ph_video.image(
        cv2.cvtColor(s['annotated'], cv2.COLOR_BGR2RGB),
        use_column_width=True,
        caption=f"Frame {s['frame_id']} | {s['fps']} FPS"
    )

    # Risk metrics
    risk_color = '🔴' if s['risk_level'] == 'HIGH' else '🟡' if s['risk_level'] == 'MEDIUM' else '🟢'
    ph_risk.metric('Risk Level', f"{risk_color} {s['risk_level']}", f"score: {s['risk_score']:.3f}")
    ph_count.metric('Pilgrims Detected', s['person_count'])
    ph_trend.metric('Trend', s['trend'].upper())

    # Arabic P0 alert
    if s.get('arabic_alert'):
        ph_alert.error(f"🚨 {s['arabic_alert']}")
        pipeline.state['arabic_alert'] = ''

    # Reflection summary
    rs = s.get('reflection_summary', {})
    if rs:
        ph_reflect.info(
            f"🪞 Reflection: {rs.get('bias_events', 0)} bias events / "
            f"{rs.get('total_reflections', 0)} frames "
            f"({rs.get('bias_rate_pct', 0):.1f}%)"
        )

cap.release()