Update app.py

app.py

@@ -1 +1,98 @@
-![image/png](https://cdn-uploads.huggingface.co/production/uploads/61b22e8eb38ed4c779d540fc/rkXD8FcVjo6w33hleoJvs.png)
+import cv2
+import gradio as gr
+from ultralytics import YOLO
+
+# ── Config ─────────────────────────────────────────────
+MODEL_PATH = "yolov8n.pt"     # modelo pre-entrenado (clase “person”)
+CONF_THRES = 0.3              # confianza mínima detección
+LINE_RATIO = 0.5              # posición de la línea virtual (50 % altura)
+# ───────────────────────────────────────────────────────
+
+model = YOLO(MODEL_PATH)
+
+# estado global
+memory = {}      # {track_id: previous_cy}
+in_count = 0
+out_count = 0
+
+
+def count_people(frame):
+    """
+    Recibe un frame RGB (numpy) -> procesa -> devuelve frame RGB anotado y string.
+    Se llama de forma continua porque el input tiene `streaming=True`.
+    """
+    global memory, in_count, out_count
+
+    if frame is None:
+        return None, ""
+
+    # ── paso 1: RGB ➜ BGR para OpenCV/YOLO ─────────────
+    frame_bgr = cv2.cvtColor(frame, cv2.COLOR_RGB2BGR)
+    h, w = frame_bgr.shape[:2]
+    line_y = int(h * LINE_RATIO)
+
+    # ── detección + tracking ───────────────────────────
+    results = model.track(
+        frame_bgr,
+        classes=[0],                 # solo “person”
+        conf=CONF_THRES,
+        persist=True,
+        verbose=False
+    )
+
+    annotated = frame_bgr.copy()
+    cv2.line(annotated, (0, line_y), (w, line_y), (0, 255, 255), 2)
+
+    if results:
+        for box in results[0].boxes:
+            x1, y1, x2, y2 = map(int, box.xyxy[0])
+            cx, cy = int((x1 + x2) / 2), int((y1 + y2) / 2)
+            tid = int(box.id[0]) if box.id is not None else -1
+
+            # cruces entrada / salida
+            prev_cy = memory.get(tid, cy)
+            if prev_cy < line_y <= cy:        # cruzó de arriba → abajo (ENTRA)
+                in_count += 1
+            elif prev_cy > line_y >= cy:      # abajo → arriba (SALE)
+                out_count += 1
+            memory[tid] = cy
+
+            # dibujitos
+            cv2.rectangle(annotated, (x1, y1), (x2, y2), (0, 255, 0), 1)
+            cv2.circle(annotated, (cx, cy), 3, (0, 0, 255), -1)
+            cv2.putText(annotated, str(tid), (x1, y1 - 5),
+                        cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 0, 0), 1)
+
+    total = in_count - out_count
+    label = f"In: {in_count} | Out: {out_count} | Ocupación: {total}"
+
+    # ── paso 2: BGR ➜ RGB para mostrar en Gradio ───────
+    annotated_rgb = cv2.cvtColor(annotated, cv2.COLOR_BGR2RGB)
+    return annotated_rgb, label
+
+
+def reset_counts():
+    """Callback para botón ‘Limpiar’."""
+    global memory, in_count, out_count
+    memory = {}
+    in_count = 0
+    out_count = 0
+    return None, ""
+
+
+with gr.Blocks(title="Contador de personas (entrada única)") as demo:
+    gr.Markdown("# Contador de personas (entrada única)")
+
+    with gr.Row():
+        cam = gr.Image(sources=["webcam"], streaming=True, label="frame")
+        out_img = gr.Image(label="Video")
+    out_lbl = gr.Text(label="Contador")
+
+    btn_clear = gr.Button("Limpiar")
+
+    # wire-up
+    cam.stream(fn=count_people, outputs=[out_img, out_lbl])
+    btn_clear.click(fn=reset_counts, outputs=[out_img, out_lbl])
+
+if __name__ == "__main__":
+    demo.launch()