🤸‍♀️ Pose Perfect AI

import cv2 import mediapipe as mp import numpy as np from typing import Dict, List, Tuple, Optional import math class PosePerfectEngine: """ Advanced pose detection and analysis engine that provides real-time feedback for fitness movements, posture correction, and movement optimization. """ def __init__(self): # Initialize MediaPipe pose detection self.mp_pose = mp.solutions.pose self.pose = self.mp_pose.Pose( static_image_mode=False, model_complexity=2, # High accuracy model enable_segmentation=True, min_detection_confidence=0.7, min_tracking_confidence=0.5 ) self.mp_drawing = mp.solutions.drawing_utils # Define key body landmarks for analysis self.key_landmarks = { 'head': [self.mp_pose.PoseLandmark.NOSE, self.mp_pose.PoseLandmark.LEFT_EAR, self.mp_pose.PoseLandmark.RIGHT_EAR], 'shoulders': [self.mp_pose.PoseLandmark.LEFT_SHOULDER, self.mp_pose.PoseLandmark.RIGHT_SHOULDER], 'arms': [self.mp_pose.PoseLandmark.LEFT_ELBOW, self.mp_pose.PoseLandmark.RIGHT_ELBOW, self.mp_pose.PoseLandmark.LEFT_WRIST, self.mp_pose.PoseLandmark.RIGHT_WRIST], 'torso': [self.mp_pose.PoseLandmark.LEFT_HIP, self.mp_pose.PoseLandmark.RIGHT_HIP], 'legs': [self.mp_pose.PoseLandmark.LEFT_KNEE, self.mp_pose.PoseLandmark.RIGHT_KNEE, self.mp_pose.PoseLandmark.LEFT_ANKLE, self.mp_pose.PoseLandmark.RIGHT_ANKLE] } # Exercise templates for form analysis self.exercise_templates = { 'squat': { 'target_angles': {'knee': 90, 'hip': 90, 'ankle': 90}, 'key_points': ['knee_alignment', 'back_straight', 'weight_distribution'] }, 'push_up': { 'target_angles': {'elbow': 90, 'shoulder': 0, 'hip': 180}, 'key_points': ['straight_line', 'elbow_position', 'full_range'] }, 'plank': { 'target_angles': {'hip': 180, 'shoulder': 90, 'ankle': 90}, 'key_points': ['neutral_spine', 'core_engagement', 'steady_position'] } } def analyze_pose_realtime(self, frame: np.ndarray, exercise_type: str = 'general') -> Dict: """ Perform real-time pose analysis on video frame. Detects landmarks, calculates angles, and provides form feedback. Args: frame: Input video frame (BGR format) exercise_type: Type of exercise being performed for specific analysis Returns: Dictionary containing pose analysis results and feedback """ # Convert BGR to RGB for MediaPipe processing rgb_frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB) # Detect pose landmarks results = self.pose.process(rgb_frame) if not results.pose_landmarks: return {'status': 'no_pose_detected', 'feedback': 'Please ensure full body is visible'} # Extract landmark coordinates landmarks = self._extract_landmark_coordinates(results.pose_landmarks, frame.shape) # Calculate body angles and measurements body_angles = self._calculate_body_angles(landmarks) # Analyze posture and alignment posture_analysis = self._analyze_posture(landmarks, body_angles) # Generate exercise-specific feedback form_feedback = self._generate_form_feedback(landmarks, body_angles, exercise_type) # Calculate overall form score form_score = self._calculate_form_score(body_angles, posture_analysis, exercise_type) # Create visualization overlay annotated_frame = self._create_pose_visualization(frame, results.pose_landmarks, form_feedback) return { 'status': 'pose_detected', 'landmarks': landmarks, 'body_angles': body_angles, 'posture_analysis': posture_analysis, 'form_feedback': form_feedback, 'form_score': form_score, 'annotated_frame': annotated_frame, 'exercise_type': exercise_type } def _calculate_body_angles(self, landmarks: Dict) -> Dict: """ Calculate important body angles for posture and movement analysis. Computes joint angles using trigonometry for accurate measurements. """ angles = {} # Calculate knee angles (important for squats, lunges) if 'left_hip' in landmarks and 'left_knee' in landmarks and 'left_ankle' in landmarks: angles['left_knee'] = self._calculate_angle( landmarks['left_hip'], landmarks['left_knee'], landmarks['left_ankle'] ) if 'right_hip' in landmarks and 'right_knee' in landmarks and 'right_ankle' in landmarks: angles['right_knee'] = self._calculate_angle( landmarks['right_hip'], landmarks['right_knee'], landmarks['right_ankle'] ) # Calculate elbow angles (important for push-ups, arm exercises) if 'left_shoulder' in landmarks and 'left_elbow' in landmarks and 'left_wrist' in landmarks: angles['left_elbow'] = self._calculate_angle( landmarks['left_shoulder'], landmarks['left_elbow'], landmarks['left_wrist'] ) if 'right_shoulder' in landmarks and 'right_elbow' in landmarks and 'right_wrist' in landmarks: angles['right_elbow'] = self._calculate_angle( landmarks['right_shoulder'], landmarks['right_elbow'], landmarks['right_wrist'] ) # Calculate hip angles (important for squats, deadlifts) if 'left_shoulder' in landmarks and 'left_hip' in landmarks and 'left_knee' in landmarks: angles['left_hip'] = self._calculate_angle( landmarks['left_shoulder'], landmarks['left_hip'], landmarks['left_knee'] ) # Calculate shoulder angles (important for overhead movements) if 'left_hip' in landmarks and 'left_shoulder' in landmarks and 'left_elbow' in landmarks: angles['left_shoulder'] = self._calculate_angle( landmarks['left_hip'], landmarks['left_shoulder'], landmarks['left_elbow'] ) return angles def _calculate_angle(self, point1: Tuple, point2: Tuple, point3: Tuple) -> float: """ Calculate angle between three points using vector mathematics. Point2 is the vertex of the angle. """ # Calculate vectors vector1 = np.array([point1[0] - point2[0], point1[1] - point2[1]]) vector2 = np.array([point3[0] - point2[0], point3[1] - point2[1]]) # Calculate angle using dot product cosine_angle = np.dot(vector1, vector2) / (np.linalg.norm(vector1) * np.linalg.norm(vector2)) # Clamp cosine value to valid range to avoid numerical errors cosine_angle = np.clip(cosine_angle, -1.0, 1.0) # Convert to degrees angle = np.arccos(cosine_angle) return np.degrees(angle) def _analyze_posture(self, landmarks: Dict, angles: Dict) -> Dict: """ Analyze overall posture and body alignment. Detects common posture issues and provides corrective guidance. """ posture_analysis = { 'overall_score': 0, 'issues': [], 'recommendations': [] } # Check shoulder alignment if 'left_shoulder' in landmarks and 'right_shoulder' in landmarks: shoulder_level = abs(landmarks['left_shoulder'][1] - landmarks['right_shoulder'][1]) if shoulder_level > 20: # Threshold for uneven shoulders posture_analysis['issues'].append('Uneven shoulders detected') posture_analysis['recommendations'].append('Focus on shoulder blade alignment') # Check hip alignment if 'left_hip' in landmarks and 'right_hip' in landmarks: hip_level = abs(landmarks['left_hip'][1] - landmarks['right_hip'][1]) if hip_level > 15: # Threshold for uneven hips posture_analysis['issues'].append('Hip misalignment detected') posture_analysis['recommendations'].append('Engage core muscles for stability') # Check spine alignment (head, shoulders, hips in line) if 'nose' in landmarks and 'left_shoulder' in landmarks and 'left_hip' in landmarks: head_forward = landmarks['nose'][0] - landmarks['left_shoulder'][0] if abs(head_forward) > 30: # Forward head posture threshold posture_analysis['issues'].append('Forward head posture') posture_analysis['recommendations'].append('Keep head aligned over shoulders') # Calculate overall posture score max_issues = 5 # Maximum possible issues current_issues = len(posture_analysis['issues']) posture_analysis['overall_score'] = max(0, (100 - (current_issues / max_issues) * 100)) return posture_analysis def _generate_form_feedback(self, landmarks: Dict, angles: Dict, exercise_type: str) -> Dict: """ Generate specific feedback based on exercise type and current form. Provides actionable corrections for optimal performance. """ feedback = { 'exercise': exercise_type, 'corrections': [], 'good_points': [], 'overall_rating': 'Good' } if exercise_type == 'squat': feedback.update(self._analyze_squat_form(landmarks, angles)) elif exercise_type == 'push_up': feedback.update(self._analyze_pushup_form(landmarks, angles)) elif exercise_type == 'plank': feedback.update(self._analyze_plank_form(landmarks, angles)) else: feedback.update(self._analyze_general_form(landmarks, angles)) return feedback def _analyze_squat_form(self, landmarks: Dict, angles: Dict) -> Dict: """ Analyze squat form with specific focus on knee tracking, depth, and balance. """ corrections = [] good_points = [] # Check squat depth using knee angle if 'left_knee' in angles: knee_angle = angles['left_knee'] if knee_angle > 120: corrections.append('Go deeper - thighs should be parallel to ground') elif knee_angle < 70: corrections.append('Avoid going too deep - maintain control') else: good_points.append('Good squat depth achieved') # Check knee alignment (knees should track over toes) if 'left_knee' in landmarks and 'left_ankle' in landmarks: knee_x = landmarks['left_knee'][0] ankle_x = landmarks['left_ankle'][0] if abs(knee_x - ankle_x) > 30: corrections.append('Keep knees aligned over toes') else: good_points.append('Good knee tracking') # Check back alignment if 'left_shoulder' in landmarks and 'left_hip' in landmarks: torso_angle = self._calculate_torso_angle(landmarks) if torso_angle < 45: corrections.append('Keep chest up and maintain neutral spine') else: good_points.append('Good torso position') rating = 'Excellent' if len(corrections) == 0 else 'Good' if len(corrections) <= 2 else 'Needs Work' return { 'corrections': corrections, 'good_points': good_points, 'overall_rating': rating }