Enhance camera stream functionality with multi-camera support

- Refactored camera handling to support multiple cameras via environment variables.
- Added API endpoints for camera selection and retrieval of available cameras.
- Updated stream processing to dynamically switch between selected cameras.
- Integrated device selection for YOLO inference to utilize CUDA if available.

camera_stream.py

@@ -3,8 +3,9 @@ import os
 import time
 import threading
 import numpy as np
+import torch
 from datetime import datetime
-from flask import Flask, Response, render_template, jsonify, send_from_directory
+from flask import Flask, Response, render_template, jsonify, send_from_directory, request
 from dotenv import load_dotenv
 from ultralytics import YOLO
 
@@ -17,16 +18,24 @@ app = Flask(__name__)
 # ---------------------------------------------------------------------------
 USERNAME = os.getenv("username")
 PASSWORD = os.getenv("password")
-RTSP_URL = (
-    f"rtsp://{USERNAME}:{PASSWORD}@192.168.6.36:554"
-    "/cam/realmonitor?channel=1&subtype=0"
-)
+CAMERA_IPS = [os.getenv(f"camera_ip_{i+1}") for i in range(1, 3)]
+CAMERAS = [
+    {
+        "id": f"cam{i + 1}",
+        "name": f"Camera {i + 1}",
+        "ip": ip,
+        "rtsp_url": f"rtsp://{USERNAME}:{PASSWORD}@{ip}:554/cam/realmonitor?channel=1&subtype=0",
+    }
+    for i, ip in enumerate(CAMERA_IPS)
+]
+DEFAULT_CAMERA_ID = CAMERAS[0]["id"] if CAMERAS else "cam1"
 
 PROXIMITY_PX = 200        # max pixel distance to consider two people "together"
 GROUP_TIME_THRESHOLD = 20  # seconds before an alert fires
 ALERT_COOLDOWN = 60        # seconds between successive alerts
 MIN_GROUP_SIZE = 2
 YOLO_CONF = 0.5
+INFERENCE_DEVICE = 0
 
 # ---------------------------------------------------------------------------
 # Shared state (protected by lock)
@@ -40,12 +49,20 @@ state = {
     "alerts": [],
     "fps": 0,
     "stream_status": "connecting",
+    "selected_camera_id": DEFAULT_CAMERA_ID,
+    "active_camera_id": DEFAULT_CAMERA_ID,
 }
 
 # ---------------------------------------------------------------------------
 # YOLO model (downloaded on first run)
 # ---------------------------------------------------------------------------
+if not torch.cuda.is_available():
+    raise RuntimeError(
+        "CUDA GPU is required but not available. Install a CUDA-enabled PyTorch build "
+        "and verify NVIDIA drivers."
+    )
 model = YOLO("yolo26m.pt")
+model.to(f"cuda:{INFERENCE_DEVICE}")
 
 # Group tracking
 _group_trackers: dict = {}
@@ -100,34 +117,70 @@ def _group_centroid(centroids, indices):
     return (sum(xs) / len(xs), sum(ys) / len(ys))
 
 
+def _get_camera_by_id(camera_id):
+    for cam in CAMERAS:
+        if cam["id"] == camera_id:
+            return cam
+    return CAMERAS[0] if CAMERAS else None
+
+
+def _reset_tracking():
+    global _group_trackers, _next_group_id, _last_alert_time
+    _group_trackers = {}
+    _next_group_id = 0
+    _last_alert_time = 0.0
+
+
 # ---------------------------------------------------------------------------
 # Main processing loop (runs in background thread)
 # ---------------------------------------------------------------------------
 
 def _process_stream():
-    global _group_trackers, _next_group_id, _last_alert_time
-
-    cap = cv2.VideoCapture(RTSP_URL)
-    if not cap.isOpened():
-        with lock:
-            state["stream_status"] = "error"
-        print(f"[ERROR] Cannot open RTSP stream: {RTSP_URL}")
-        return
-
-    with lock:
-        state["stream_status"] = "live"
-
+    global _next_group_id, _last_alert_time
+    cap = None
+    active_camera_id = None
     prev_time = time.time()
 
     while True:
+        with lock:
+            selected_camera_id = state["selected_camera_id"]
+        selected_camera = _get_camera_by_id(selected_camera_id)
+        if not selected_camera:
+            with lock:
+                state["stream_status"] = "error"
+            time.sleep(2)
+            continue
+
+        if cap is None or active_camera_id != selected_camera["id"]:
+            if cap is not None:
+                cap.release()
+            cap = cv2.VideoCapture(selected_camera["rtsp_url"])
+            active_camera_id = selected_camera["id"]
+            _reset_tracking()
+            with lock:
+                state["active_camera_id"] = active_camera_id
+            if not cap.isOpened():
+                with lock:
+                    state["stream_status"] = "error"
+                print(f"[ERROR] Cannot open RTSP stream: {selected_camera['rtsp_url']}")
+                time.sleep(2)
+                cap = None
+                continue
+            with lock:
+                state["stream_status"] = "live"
+
         ret, frame = cap.read()
         if not ret:
             with lock:
                 state["stream_status"] = "reconnecting"
             cap.release()
             time.sleep(2)
-            cap = cv2.VideoCapture(RTSP_URL)
-            if cap.isOpened():
+            cap = cv2.VideoCapture(selected_camera["rtsp_url"])
+            if not cap.isOpened():
+                with lock:
+                    state["stream_status"] = "error"
+                cap = None
+            else:
                 with lock:
                     state["stream_status"] = "live"
             continue
@@ -137,7 +190,13 @@ def _process_stream():
         prev_time = now
 
         # --- YOLO inference (person = class 0) ---
-        results = model(frame, classes=[0], verbose=False, conf=YOLO_CONF)
+        results = model(
+            frame,
+            classes=[0],
+            verbose=False,
+            conf=YOLO_CONF,
+            device=INFERENCE_DEVICE,
+        )
 
         person_boxes = []
         for r in results:
@@ -309,6 +368,7 @@ def video_feed():
 @app.route("/api/status")
 def api_status():
     with lock:
+        active_camera = _get_camera_by_id(state["active_camera_id"])
         return jsonify({
             "people_count": state["people_count"],
             "groups": state["groups"],
@@ -316,9 +376,36 @@ def api_status():
             "alerts": state["alerts"][:20],
             "fps": state["fps"],
             "stream_status": state["stream_status"],
+            "selected_camera_id": state["selected_camera_id"],
+            "active_camera_name": active_camera["name"] if active_camera else "Unknown",
         })
 
 
+@app.route("/api/cameras")
+def api_cameras():
+    return jsonify({
+        "cameras": [{"id": c["id"], "name": c["name"], "ip": c["ip"]} for c in CAMERAS]
+    })
+
+
+@app.route("/api/camera/select", methods=["POST"])
+def api_camera_select():
+    data = request.get_json(silent=True) or {}
+    camera_id = data.get("camera_id")
+    camera = _get_camera_by_id(camera_id)
+    if camera is None:
+        return jsonify({"ok": False, "error": "Invalid camera id"}), 400
+    with lock:
+        state["selected_camera_id"] = camera["id"]
+        state["frame"] = None
+        state["groups"] = []
+        state["people_count"] = 0
+        state["alert_active"] = False
+        state["fps"] = 0
+        state["stream_status"] = "connecting"
+    return jsonify({"ok": True})
+
+
 @app.route("/alerts/<path:filename>")
 def serve_alert_image(filename):
     return send_from_directory("alerts", filename)