Update Pose_And_Motion_Based_SVM_Model.py

pages/Pose_And_Motion_Based_SVM_Model.py

@@ -17,11 +17,69 @@ ocsvm_pose = joblib.load('pages/ocsvm_pose_model-4.pkl')  # Pre-trained One-Clas
 
 #==========================FUNCTIONS==========================
 
-
+# Function to initialize OpenPose
+def initialize_openpose():
+    params = {
+        "model_folder": "openpose/models",  # Path to OpenPose models folder
+        "hand": True,  # If you want hand keypoints
+        "face": False,  # If you want face keypoints (set to True if needed)
+        "body": 25  # Number of body keypoints (e.g., 18 or 25 depending on your model)
+    }
+    opWrapper = op.WrapperPython()
+    opWrapper.configure(params)
+    opWrapper.start()
+    return opWrapper
 
 # Function to extract poses from the video
 def extract_poses(video_path):
-    test = 'hello'
+    # Initialize OpenPose
+    opWrapper = initialize_openpose()
+
+    # Open the video file
+    cap = cv2.VideoCapture(video_path)
+    frame_count = 0
+    all_pose_data = []
+
+    while cap.isOpened():
+        ret, frame = cap.read()
+        if not ret:
+            break  # End of video
+
+        # Extract poses for the current frame
+        datum = op.Datum()
+        datum.cvInputData = frame
+        opWrapper.emplaceAndPop([datum])
+
+        # Extract the body keypoints (datum.poseKeypoints is of shape (num_people, 25, 3) if using 25 keypoints)
+        pose_keypoints = datum.poseKeypoints
+
+        if pose_keypoints is not None:
+            # Extract features for each person (you can customize this)
+            for person_keypoints in pose_keypoints:
+                # Flatten the keypoints to a single vector for each person
+                flat_keypoints = person_keypoints.flatten()
+                all_pose_data.append(flat_keypoints)
+        
+        frame_count += 1
+
+    cap.release()
+
+    # If pose data is collected, predict with One-Class SVM
+    if all_pose_data:
+        pose_data = np.array(all_pose_data)
+
+        # Preprocess pose data (scale using the same scaler used during training)
+        pose_data_scaled = pose_scaler.transform(pose_data)
+
+        # Make predictions with the One-Class SVM
+        predictions = ocsvm_pose.predict(pose_data_scaled)
+        
+        # Show predictions (for demonstration purposes, you can adjust as needed)
+        st.write(f"Predictions (1 = normal, -1 = panic): {predictions}")
+        
+        # Optionally, process the predictions (e.g., count the number of panic poses)
+        panic_count = np.sum(predictions == -1)
+        st.write(f"Number of panic poses detected: {panic_count}")
 
 #==========================PREDICTING UPLOADED VIDEO==========================