Update app.py

app.py

@@ -2,76 +2,68 @@ from fastapi import FastAPI, UploadFile, File
 import cv2
 import numpy as np
 import mediapipe as mp
-import torch
-from ultralytics import YOLO
 
-app = FastAPI()
+app = FastAPI(title="Human Anthropometry API")
 
-# Load models
-yolo = YOLO("yolov8n.pt")
-mp_pose = mp.solutions.pose.Pose(static_image_mode=True)
+mp_pose = mp.solutions.pose
+pose = mp_pose.Pose(static_image_mode=False)
 
-# ---------------- UTILITIES ---------------- #
+# ---- Utility ----
+def dist(a, b):
+    return np.linalg.norm(a - b)
 
-def distance(p1, p2):
-    return np.linalg.norm(np.array(p1) - np.array(p2))
+def analyze_frame(image):
+    h, w, _ = image.shape
+    rgb = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
+    res = pose.process(rgb)
 
-def estimate_scale(bbox_height_px, camera_fov=60):
-    # empirical scale estimation
-    return 170 / bbox_height_px
+    if not res.pose_landmarks:
+        return None
 
-# ---------------- CORE LOGIC ---------------- #
+    lm = res.pose_landmarks.landmark
 
-@app.post("/analyze-image")
-async def analyze_image(file: UploadFile = File(...)):
-    image_bytes = await file.read()
-    img_np = np.frombuffer(image_bytes, np.uint8)
-    image = cv2.imdecode(img_np, cv2.IMREAD_COLOR)
+    def p(i):
+        return np.array([lm[i].x * w, lm[i].y * h])
 
-    # 1. Person detection
-    results = yolo(image, conf=0.4)
-    if len(results[0].boxes) == 0:
-        return {"error": "No human detected"}
+    head = p(0)
+    l_sh = p(11)
+    r_sh = p(12)
+    l_hip = p(23)
+    r_hip = p(24)
+    l_heel = p(29)
+    r_heel = p(30)
 
-    box = results[0].boxes.xyxy[0].cpu().numpy()
-    x1, y1, x2, y2 = map(int, box)
-    person = image[y1:y2, x1:x2]
+    shoulder_px = dist(l_sh, r_sh)
+    torso_px = dist((l_sh + r_sh)/2, (l_hip + r_hip)/2)
+    height_px = dist(head, (l_heel + r_heel)/2)
 
-    # 2. Pose estimation
-    rgb = cv2.cvtColor(person, cv2.COLOR_BGR2RGB)
-    pose_result = mp_pose.process(rgb)
-    if not pose_result.pose_landmarks:
-        return {"error": "Pose not detected"}
+    # ---- INDUSTRIAL NORMALIZATION ----
+    # Average adult shoulder width ≈ 0.23 × height
+    height_cm = (shoulder_px / height_px) * (1 / 0.23) * 100
 
-    h, w, _ = person.shape
-    lm = pose_result.pose_landmarks.landmark
+    shoulder_cm = height_cm * 0.23
+    torso_ratio = torso_px / height_px
 
-    left_shoulder = (lm[11].x * w, lm[11].y * h)
-    right_shoulder = (lm[12].x * w, lm[12].y * h)
-    left_hip = (lm[23].x * w, lm[23].y * h)
-    right_hip = (lm[24].x * w, lm[24].y * h)
+    # Weight estimation via body volume proxy (not BMI)
+    weight_kg = round((height_cm * shoulder_cm * torso_ratio) / 1000, 1)
 
-    shoulder_px = distance(left_shoulder, right_shoulder)
-    hip_px = distance(left_hip, right_hip)
+    confidence = round(min(0.95, 0.6 + torso_ratio), 2)
 
-    # 3. Scale estimation
-    bbox_height = y2 - y1
-    scale = estimate_scale(bbox_height)
+    return {
+        "height_cm": round(height_cm, 1),
+        "shoulder_cm": round(shoulder_cm, 1),
+        "weight_kg": weight_kg,
+        "confidence": confidence
+    }
 
-    # 4. Final metrics
-    height_cm = bbox_height * scale
-    shoulder_cm = shoulder_px * scale
+# ---- API ----
+@app.post("/analyze")
+async def analyze(file: UploadFile = File(...)):
+    data = await file.read()
+    img = cv2.imdecode(np.frombuffer(data, np.uint8), cv2.IMREAD_COLOR)
 
-    # Learned-like regression (approx industrial)
-    weight_kg = (
-        0.45 * height_cm +
-        1.2 * shoulder_cm +
-        0.8 * hip_px * scale
-    ) / 10
+    result = analyze_frame(img)
+    if not result:
+        return {"error": "Human not detected"}
 
-    return {
-        "height_cm": round(height_cm, 2),
-        "shoulder_cm": round(shoulder_cm, 2),
-        "weight_kg": round(weight_kg, 2),
-        "confidence": 0.9
-    }
+    return result