Update app.py

app.py

@@ -2,162 +2,184 @@ import gradio as gr
 import cv2
 import numpy as np
 import mediapipe as mp
+import torch
+import torch.nn as nn
 import tempfile
 import os
 import math
 
-# تهيئة MediaPipe Pose
+# ===========================
+# نموذج تصنيف المشية
+# ===========================
+class SimpleGaitClassifier(nn.Module):
+    def __init__(self, input_dim, hidden_dim=128, num_classes=2):
+        super().__init__()
+        self.net = nn.Sequential(
+            nn.Linear(input_dim, hidden_dim),
+            nn.ReLU(),
+            nn.Linear(hidden_dim, hidden_dim),
+            nn.ReLU(),
+            nn.Linear(hidden_dim, num_classes)
+        )
+    def forward(self, x):
+        return self.net(x)
+
+input_dim = 33*3  # 33 نقطة لكل من x,y,z
+model = SimpleGaitClassifier(input_dim)
+model.eval()  # وضع النموذج في وضع الاختبار
+
+classes = ['Normal', 'Abnormal']
+
+# ===========================
+# MediaPipe لاستخراج النقاط
+# ===========================
 mp_pose = mp.solutions.pose
 pose = mp_pose.Pose(static_image_mode=False, min_detection_confidence=0.7, model_complexity=1)
 
-def calculate_distance(p1, p2):
-    return math.sqrt((p1[0]-p2[0])**2 + (p1[1]-p2[1])**2)
-
-def calculate_angle(a, b, c):
-    try:
-        a, b, c = np.array(a), np.array(b), np.array(c)
-        radians = np.arctan2(c[1]-b[1], c[0]-b[0]) - np.arctan2(a[1]-b[1], a[0]-b[0])
-        angle = abs(radians * 180.0 / np.pi)
-        return angle if angle <= 180 else 360 - angle
-    except:
-        return 0
-
-def analyze_gait(video_file):
-    if video_file is None:
-        return "❌ يرجى رفع فيديو أولاً"
-    
-    # التعامل مع File من Gradio
-    if hasattr(video_file, 'name'):
-        video_path = video_file.name
-    else:
-        temp_video = tempfile.NamedTemporaryFile(delete=False, suffix=".mp4")
-        with open(temp_video.name, 'wb') as f:
-            f.write(video_file)
-        video_path = temp_video.name
-        temp_video.close()
-
+def extract_keypoints_from_video(video_path, max_frames=100):
     cap = cv2.VideoCapture(video_path)
-    if not cap.isOpened():
-        return "❌ لا يمكن فتح الفيديو"
-
-    total_frames = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
-    fps = cap.get(cv2.CAP_PROP_FPS)
-    frame_width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
-    frame_height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
-    pixel_to_meter = 1.7 / (frame_height * 0.8)  # تقدير طول الشخص ~1.7 م
-
-    # تخزين القياسات
-    left_clearances, right_clearances = [], []
-    left_angles, right_angles = [], []
-    base_widths = []
-    left_steps, right_steps = [], []
-    prev_left_ankle, prev_right_ankle = None, None
+    keypoints_seq = []
     frames_processed = 0
-    person_detected = False
-
-    while cap.isOpened() and frames_processed < min(500, total_frames):
+    while cap.isOpened() and frames_processed < max_frames:
         ret, frame = cap.read()
         if not ret:
             break
-
         frame_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
         results = pose.process(frame_rgb)
         if results.pose_landmarks:
-            person_detected = True
             lm = results.pose_landmarks.landmark
-
-            # نقاط رئيسية
-            left_ankle = [lm[mp_pose.PoseLandmark.LEFT_ANKLE].x*frame_width, lm[mp_pose.PoseLandmark.LEFT_ANKLE].y*frame_height]
-            right_ankle = [lm[mp_pose.PoseLandmark.RIGHT_ANKLE].x*frame_width, lm[mp_pose.PoseLandmark.RIGHT_ANKLE].y*frame_height]
-            left_knee = [lm[mp_pose.PoseLandmark.LEFT_KNEE].x*frame_width, lm[mp_pose.PoseLandmark.LEFT_KNEE].y*frame_height]
-            right_knee = [lm[mp_pose.PoseLandmark.RIGHT_KNEE].x*frame_width, lm[mp_pose.PoseLandmark.RIGHT_KNEE].y*frame_height]
-            left_heel = [lm[mp_pose.PoseLandmark.LEFT_HEEL].x*frame_width, lm[mp_pose.PoseLandmark.LEFT_HEEL].y*frame_height]
-            right_heel = [lm[mp_pose.PoseLandmark.RIGHT_HEEL].x*frame_width, lm[mp_pose.PoseLandmark.RIGHT_HEEL].y*frame_height]
-            left_foot = [lm[mp_pose.PoseLandmark.LEFT_FOOT_INDEX].x*frame_width, lm[mp_pose.PoseLandmark.LEFT_FOOT_INDEX].y*frame_height]
-            right_foot = [lm[mp_pose.PoseLandmark.RIGHT_FOOT_INDEX].x*frame_width, lm[mp_pose.PoseLandmark.RIGHT_FOOT_INDEX].y*frame_height]
-
-            # ارتفاع القدم
-            ground_threshold = frame_height*0.92
-            left_clear = max(0, (ground_threshold - min(left_ankle[1], left_foot[1]))*pixel_to_meter*100)
-            right_clear = max(0, (ground_threshold - min(right_ankle[1], right_foot[1]))*pixel_to_meter*100)
-            left_clearances.append(left_clear)
-            right_clearances.append(right_clear)
-
-            # زاوية الكاحل
-            left_angles.append(calculate_angle(left_knee, left_ankle, left_foot))
-            right_angles.append(calculate_angle(right_knee, right_ankle, right_foot))
-
-            # عرض القاعدة
-            base_widths.append(abs(left_ankle[0]-right_ankle[0])*pixel_to_meter)
-
-            # طول الخطوة
-            if prev_left_ankle: left_steps.append(calculate_distance(left_ankle, prev_left_ankle)*pixel_to_meter)
-            if prev_right_ankle: right_steps.append(calculate_distance(right_ankle, prev_right_ankle)*pixel_to_meter)
-
-            prev_left_ankle = left_ankle
-            prev_right_ankle = right_ankle
+            keypoints = []
+            for point in lm:
+                keypoints.extend([point.x, point.y, point.z])
+            keypoints_seq.append(keypoints)
             frames_processed += 1
-
     cap.release()
-    try: os.unlink(video_path)
-    except: pass
-
-    if not person_detected:
-        return "❌ لم يتم اكتشاف شخص في الفيديو"
-
-    # إحصاءات دقيقة
-    avg_left_clear, avg_right_clear = np.mean(left_clearances), np.mean(right_clearances)
-    std_left_clear, std_right_clear = np.std(left_clearances), np.std(right_clearances)
-    avg_left_angle, avg_right_angle = np.mean(left_angles), np.mean(right_angles)
-    std_left_angle, std_right_angle = np.std(left_angles), np.std(right_angles)
-    avg_base = np.mean(base_widths)
-    avg_left_step, avg_right_step = np.mean(left_steps) if left_steps else 0, np.mean(right_steps) if right_steps else 0
-
-    # تحديد الجانب المتضرر
-    if (avg_left_clear < avg_right_clear) or (avg_left_angle < avg_right_angle):
-        affected_side = "اليسار"
+    return keypoints_seq if keypoints_seq else [[0]*input_dim]
+
+# ===========================
+# تحليل المشية بعد تصنيفها
+# ===========================
+def analyze_gait_with_classifier(video_file):
+    if video_file is None:
+        return "❌ يرجى رفع فيديو أولاً"
+    
+    # حفظ الفيديو مؤقتًا
+    if hasattr(video_file, 'name'):
+        video_path = video_file.name
     else:
-        affected_side = "اليمين"
+        temp_video = tempfile.NamedTemporaryFile(delete=False, suffix=".mp4")
+        with open(temp_video.name, 'wb') as f:
+            f.write(video_file)
+        video_path = temp_video.name
+        temp_video.close()
 
+    # استخراج النقاط
+    keypoints_seq = extract_keypoints_from_video(video_path)
+    x = torch.tensor(keypoints_seq, dtype=torch.float32).mean(dim=0).unsqueeze(0)
+    
     # تصنيف المشية
-    gait_issue = False
-    condition, severity = "غير محددة", "طبيعية"
-
-    # معايير دقيقة للكشف عن المشكلات
-    if (avg_left_clear < 8 or avg_right_clear < 8) or abs(avg_left_angle - avg_right_angle) > 10 or avg_left_step < 0.3 or avg_right_step < 0.3:
-        gait_issue = True
-        if avg_base > 0.22: condition = "قدم شاركوت (Charcot Foot)"; severity = "خطر شديد"
-        elif avg_left_clear < 8 or avg_right_clear < 8: condition = "ضعف Tibialis Anterior / Foot Drop"; severity = "معتدل"
-        else: condition = "اعتلال الأعصاب المحيطية"; severity = "معتدل"
-
-    # التقرير النهائي
-    if not gait_issue:
-        report = f"✅ المشية طبيعية. لا توجد علامات واضحة على المشكلات.\n\n**ملاحظة:** هذا تحليل مبدئي ولا يمكن الاعتماد عليه وحده لتحديد المشكلة."
+    with torch.no_grad():
+        output = model(x)
+        pred_class = torch.argmax(output, dim=1).item()
+    
+    gait_label = classes[pred_class]
+    
+    if gait_label == 'Normal':
+        report = "✅ المشية طبيعية. لا توجد علامات واضحة على المشكلات.\n\n**ملاحظة:** هذا تحليل مبدئي ولا يمكن الاعتماد عليه وحده لتحديد المشكلة."
     else:
+        # تحليل أساسي لتحديد الجانب المتضرر والحالة
+        # نفس فكرة ارتفاع القدم وزوايا الكاحل
+        cap = cv2.VideoCapture(video_path)
+        frame_width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
+        frame_height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
+        pixel_to_meter = 1.7 / (frame_height*0.8)
+        
+        left_clearances, right_clearances = [], []
+        left_angles, right_angles = [], []
+        prev_left_ankle, prev_right_ankle = None, None
+        left_steps, right_steps = [], []
+        frames_processed = 0
+        person_detected = False
+        
+        while cap.isOpened() and frames_processed < min(500, int(cap.get(cv2.CAP_PROP_FRAME_COUNT))):
+            ret, frame = cap.read()
+            if not ret:
+                break
+            frame_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
+            results = pose.process(frame_rgb)
+            if results.pose_landmarks:
+                person_detected = True
+                lm = results.pose_landmarks.landmark
+                left_ankle = [lm[mp_pose.PoseLandmark.LEFT_ANKLE].x*frame_width, lm[mp_pose.PoseLandmark.LEFT_ANKLE].y*frame_height]
+                right_ankle = [lm[mp_pose.PoseLandmark.RIGHT_ANKLE].x*frame_width, lm[mp_pose.PoseLandmark.RIGHT_ANKLE].y*frame_height]
+                left_knee = [lm[mp_pose.PoseLandmark.LEFT_KNEE].x*frame_width, lm[mp_pose.PoseLandmark.LEFT_KNEE].y*frame_height]
+                right_knee = [lm[mp_pose.PoseLandmark.RIGHT_KNEE].x*frame_width, lm[mp_pose.PoseLandmark.RIGHT_KNEE].y*frame_height]
+                left_heel = [lm[mp_pose.PoseLandmark.LEFT_HEEL].x*frame_width, lm[mp_pose.PoseLandmark.LEFT_HEEL].y*frame_height]
+                right_heel = [lm[mp_pose.PoseLandmark.RIGHT_HEEL].x*frame_width, lm[mp_pose.PoseLandmark.RIGHT_HEEL].y*frame_height]
+                left_foot = [lm[mp_pose.PoseLandmark.LEFT_FOOT_INDEX].x*frame_width, lm[mp_pose.PoseLandmark.LEFT_FOOT_INDEX].y*frame_height]
+                right_foot = [lm[mp_pose.PoseLandmark.RIGHT_FOOT_INDEX].x*frame_width, lm[mp_pose.PoseLandmark.RIGHT_FOOT_INDEX].y*frame_height]
+
+                ground_threshold = frame_height*0.92
+                left_clear = max(0,(ground_threshold - min(left_ankle[1], left_foot[1]))*pixel_to_meter*100)
+                right_clear = max(0,(ground_threshold - min(right_ankle[1], right_foot[1]))*pixel_to_meter*100)
+                left_clearances.append(left_clear)
+                right_clearances.append(right_clear)
+
+                def calc_angle(a,b,c):
+                    a,b,c = np.array(a), np.array(b), np.array(c)
+                    angle = abs(math.degrees(math.atan2(c[1]-b[1],c[0]-b[0])-math.atan2(a[1]-b[1],a[0]-b[0])))
+                    return angle if angle<=180 else 360-angle
+
+                left_angles.append(calc_angle(left_knee,left_ankle,left_foot))
+                right_angles.append(calc_angle(right_knee,right_ankle,right_foot))
+
+                if prev_left_ankle: left_steps.append(calculate_distance(left_ankle, prev_left_ankle)*pixel_to_meter)
+                if prev_right_ankle: right_steps.append(calculate_distance(right_ankle, prev_right_ankle)*pixel_to_meter)
+                prev_left_ankle, prev_right_ankle = left_ankle, right_ankle
+                frames_processed += 1
+        
+        cap.release()
+        try: os.unlink(video_path)
+        except: pass
+        
+        avg_left_clear, avg_right_clear = np.mean(left_clearances), np.mean(right_clearances)
+        avg_left_angle, avg_right_angle = np.mean(left_angles), np.mean(right_angles)
+        avg_left_step, avg_right_step = np.mean(left_steps) if left_steps else 0, np.mean(right_steps) if right_steps else 0
+
+        affected_side = "اليسار" if avg_left_clear<avg_right_clear or avg_left_angle<avg_right_angle else "اليمين"
+
+        if avg_left_clear<8 or avg_right_clear<8:
+            condition="ضعف Tibialis Anterior / Foot Drop"
+        elif abs(avg_left_angle-avg_right_angle)>10 or avg_left_step<0.3 or avg_right_step<0.3:
+            condition="اعتلال الأعصاب المحيطية"
+        else:
+            condition="قدم شاركوت (Charcot Foot)"
+
         report = f"""⚠️ تم اكتشاف مشكلات في المشية
 - الجانب المتأثر: **{affected_side}**
-- شدة الحالة: **{severity}**
 - الحالة المحتملة: **{condition}**
 - نوصي بمراجعة طبيب مختص.
 - يمكن حجز موعد مباشر **حضوري أو أونلاين** عبر التطبيق.
 ---
 **ملاحظة:** هذا تحليل مبدئي ولا يمكن الاعتماد عليه وحده لتحديد المشكلة.
 """
+
     return report
 
-# واجهة Gradio محسنة
-with gr.Blocks(title="تحليل المشية العصبية المحسّن") as demo:
-    gr.Markdown("# 🩺 تحليل المشية العصبية المحسّن")
-    gr.Markdown("رفع فيديو المشي للحصول على تقييم مبدئي للحالة.")
-    
+# ===========================
+# واجهة Gradio
+# ===========================
+with gr.Blocks(title="تصنيف المشية العصبية") as demo:
+    gr.Markdown("# 🩺 تحليل المشية العصبية باستخدام نموذج بسيط")
+    gr.Markdown("رفع فيديو المشي للحصول على تقييم مبدئي.")
     with gr.Row():
         with gr.Column(scale=1):
             video_input = gr.File(label="اختر ملف الفيديو", file_types=[".mp4",".avi",".mov"], type="binary")
-            analyze_btn = gr.Button("بدء التحليل", variant="primary")
+            analyze_btn = gr.Button("تحليل المشية", variant="primary")
         with gr.Column(scale=2):
             output_report = gr.Markdown(value="**سيظهر التقرير هنا**")
     
-    analyze_btn.click(fn=analyze_gait, inputs=[video_input], outputs=[output_report])
+    analyze_btn.click(fn=analyze_gait_with_classifier, inputs=[video_input], outputs=[output_report])
 
 if __name__ == "__main__":
     demo.launch(server_name="0.0.0.0", server_port=7860)