app.py

import os
import math
import numpy as np
import cv2
from collections import deque
from ultralytics import YOLO
import mediapipe as mp
from flask import Flask, render_template, request, send_file, jsonify, Response
from werkzeug.utils import secure_filename
import zipfile
import shutil
import json
from queue import Queue

app = Flask(__name__)
app.config['UPLOAD_FOLDER'] = 'uploads'
app.config['OUTPUT_FOLDER'] = 'outputs'
app.config['MAX_CONTENT_LENGTH'] = 500 * 1024 * 1024  # 500MB max file size

# Create necessary folders
os.makedirs(app.config['UPLOAD_FOLDER'], exist_ok=True)
os.makedirs(app.config['OUTPUT_FOLDER'], exist_ok=True)

# Progress tracking
progress_queue = Queue()

# Load YOLO model
model = YOLO('yolov8n.pt')

# MediaPipe setup
mp_drawing = mp.solutions.drawing_utils
mp_pose = mp.solutions.pose

def calculate_angle(shoulder_center, hip_center):
    dy = shoulder_center[1] - hip_center[1]
    dx = shoulder_center[0] - hip_center[0]
    angle = math.atan2(dy, dx)
    return abs(90 - np.degrees(angle))

def classify_posture(torso_angle, standing_threshold=20, horizontal_threshold=50):
    if torso_angle < standing_threshold:
        return "Standing"
    elif torso_angle > horizontal_threshold:
        return "Lying"
    else:
        return "Falling"

def process_video(video_path, output_folder):
    camera = cv2.VideoCapture(video_path)
    fps = camera.get(cv2.CAP_PROP_FPS)
    width = int(camera.get(cv2.CAP_PROP_FRAME_WIDTH))
    height = int(camera.get(cv2.CAP_PROP_FRAME_HEIGHT))
    total_frames = int(camera.get(cv2.CAP_PROP_FRAME_COUNT))
    
    # Process every N frames to speed up (CPU is slow)
    PROCESS_EVERY_N_FRAMES = 5  # 5x faster - process 1 out of every 5 frames
    
    output_video_path = os.path.join(output_folder, 'processed_video.mp4')
    # Use mp4v codec - most compatible with OpenCV without extra dependencies
    fourcc = cv2.VideoWriter_fourcc(*'mp4v')
    video_writer = cv2.VideoWriter(output_video_path, fourcc, fps, (width, height))
    
    if not video_writer.isOpened():
        print(f"ERROR: Failed to initialize video writer with mp4v codec")
        print(f"Video properties: {width}x{height} @ {fps} fps")
        raise Exception("Failed to initialize video writer")
    
    fall_counter = 0
    fall_detected = False
    fall_video_writer = None
    fall_video_count = 0
    frames_after_fall = 0
    frames_after_fall_duration = 10
    
    buffer_size = 10
    frame_buffer = deque(maxlen=buffer_size)
    
    fall_videos = []
    frame_count = 0
    
    with mp_pose.Pose(min_detection_confidence=0.5, min_tracking_confidence=0.5) as pose:
        while camera.isOpened():
            ret, frame = camera.read()
            if not ret:
                break
            
            frame_count += 1
            
            # Report progress
            progress = int((frame_count / total_frames) * 100)
            progress_queue.put(progress)
            
            # Skip frames for speed
            if frame_count % PROCESS_EVERY_N_FRAMES != 0:
                frame_buffer.append(frame)
                video_writer.write(frame)
                if fall_video_writer is not None:
                    fall_video_writer.write(frame)
                continue
            
            results = model(frame)
            for result in results:
                for bbox, cls in zip(result.boxes.xyxy, result.boxes.cls):
                    if int(cls) == 0:  # Person class
                        x1, y1, x2, y2 = map(int, bbox)
                        person_bbox = frame[y1:y2, x1:x2]
                        
                        person_bbox_rgb = cv2.cvtColor(person_bbox, cv2.COLOR_BGR2RGB)
                        person_results = pose.process(person_bbox_rgb)
                        
                        if person_results.pose_landmarks:
                            landmarks = person_results.pose_landmarks.landmark
                            shoulders = [
                                (landmarks[mp_pose.PoseLandmark.LEFT_SHOULDER.value].x * person_bbox.shape[1],
                                 landmarks[mp_pose.PoseLandmark.LEFT_SHOULDER.value].y * person_bbox.shape[0]),
                                (landmarks[mp_pose.PoseLandmark.RIGHT_SHOULDER.value].x * person_bbox.shape[1],
                                 landmarks[mp_pose.PoseLandmark.RIGHT_SHOULDER.value].y * person_bbox.shape[0])
                            ]
                            hips = [
                                (landmarks[mp_pose.PoseLandmark.LEFT_HIP.value].x * person_bbox.shape[1],
                                 landmarks[mp_pose.PoseLandmark.LEFT_HIP.value].y * person_bbox.shape[0]),
                                (landmarks[mp_pose.PoseLandmark.RIGHT_HIP.value].x * person_bbox.shape[1],
                                 landmarks[mp_pose.PoseLandmark.RIGHT_HIP.value].y * person_bbox.shape[0])
                            ]
                            
                            shoulder_center = ((shoulders[0][0] + shoulders[1][0]) / 2,
                                             (shoulders[0][1] + shoulders[1][1]) / 2)
                            hip_center = ((hips[0][0] + hips[1][0]) / 2,
                                        (hips[0][1] + hips[1][1]) / 2)
                            
                            torso_angle = calculate_angle(hip_center, shoulder_center)
                            posture = classify_posture(torso_angle)
                            
                            if posture == "Falling":
                                fall_counter += 1
                                frames_after_fall = 0
                                if fall_counter >= 0.1 * fps and fall_counter <= 0.5 * fps and fall_video_writer is None:
                                    fall_video_count += 1
                                    fall_video_file = os.path.join(output_folder, f'fall_{fall_video_count}.mp4')
                                    fall_video_writer = cv2.VideoWriter(fall_video_file, cv2.VideoWriter_fourcc(*'mp4v'), fps, (width, height))
                                    fall_videos.append(fall_video_file)
                                    
                                    while frame_buffer:
                                        fall_video_writer.write(frame_buffer.popleft())
                            else:
                                fall_counter = 0
                            
                            if fall_video_writer is not None:
                                fall_video_writer.write(frame)
                            
                            if posture == "Standing" and fall_video_writer is not None:
                                frames_after_fall += 1
                        
                        frame[y1:y2, x1:x2] = person_bbox
            
            frame_buffer.append(frame)
            video_writer.write(frame)
            
            if frames_after_fall > frames_after_fall_duration and fall_video_writer is not None:
                fall_video_writer.release()
                fall_video_writer = None
    
    camera.release()
    video_writer.release()
    
    if fall_video_writer is not None:
        fall_video_writer.release()
    
    # Check if video was created successfully
    if os.path.exists(output_video_path) and os.path.getsize(output_video_path) > 0:
        print(f"SUCCESS: Processed video created: {output_video_path} ({os.path.getsize(output_video_path)} bytes)")
    else:
        print(f"ERROR: Processed video not created or empty: {output_video_path}")
        raise Exception("Failed to create processed video")
    
    return output_video_path, fall_videos

@app.route('/')
def index():
    return render_template('index.html')

@app.route('/progress')
def progress():
    def generate():
        last_progress = 0
        while True:
            try:
                progress = progress_queue.get(timeout=0.5)
                last_progress = progress
                yield f"data: {json.dumps({'progress': progress})}\n\n"
                if progress >= 100:
                    break
            except:
                # Keep sending last known progress instead of 0
                if last_progress < 100:
                    yield f"data: {json.dumps({'progress': last_progress})}\n\n"
                else:
                    break
    
    return Response(generate(), mimetype='text/event-stream')

@app.route('/upload', methods=['POST'])
def upload_file():
    if 'video' not in request.files:
        return jsonify({'error': 'No video file uploaded'}), 400
    
    file = request.files['video']
    if file.filename == '':
        return jsonify({'error': 'No file selected'}), 400
    
    if file:
        # Clear progress queue
        while not progress_queue.empty():
            try:
                progress_queue.get_nowait()
            except:
                break
        
        # Clean previous uploads and outputs
        for folder in [app.config['UPLOAD_FOLDER'], app.config['OUTPUT_FOLDER']]:
            for filename in os.listdir(folder):
                file_path = os.path.join(folder, filename)
                try:
                    if os.path.isfile(file_path):
                        os.unlink(file_path)
                except Exception as e:
                    print(f"Error deleting {file_path}: {e}")
        
        filename = secure_filename(file.filename)
        filepath = os.path.join(app.config['UPLOAD_FOLDER'], filename)
        file.save(filepath)
        
        try:
            processed_video, fall_videos = process_video(filepath, app.config['OUTPUT_FOLDER'])
            
            # Create zip file with all outputs
            zip_path = os.path.join(app.config['OUTPUT_FOLDER'], 'results.zip')
            print(f"Creating ZIP file: {zip_path}")
            with zipfile.ZipFile(zip_path, 'w') as zipf:
                zipf.write(processed_video, 'processed_video.mp4')
                print(f"Added to ZIP: processed_video.mp4")
                for i, fall_video in enumerate(fall_videos, 1):
                    zipf.write(fall_video, f'fall_{i}.mp4')
                    print(f"Added to ZIP: fall_{i}.mp4")
            
            print(f"ZIP created successfully. Fall count: {len(fall_videos)}")
            return jsonify({
                'success': True,
                'fall_count': len(fall_videos),
                'download_url': '/download'
            })
        except Exception as e:
            print(f"ERROR in upload: {str(e)}")
            import traceback
            traceback.print_exc()
            return jsonify({'error': str(e)}), 500

@app.route('/download')
def download_file():
    zip_path = os.path.join(app.config['OUTPUT_FOLDER'], 'results.zip')
    return send_file(zip_path, as_attachment=True, download_name='fall_detection_results.zip')

@app.route('/video/<filename>')
def serve_video(filename):
    video_path = os.path.join(app.config['OUTPUT_FOLDER'], filename)
    response = send_file(video_path, mimetype='video/mp4')
    response.headers['Accept-Ranges'] = 'bytes'
    response.headers['Cache-Control'] = 'no-cache'
    return response

if __name__ == '__main__':
    app.run(host='0.0.0.0', port=7860, debug=False)