Stranger Detection System - Video Processing Pipeline

This is a computer vision project for face analysis from surveillance videos. Currently includes STEP-1 through STEP-4.

STEP-1: Video Dataset Preparation and Access

This step focuses on video dataset preparation and access.

Usage for STEP-1

1. Place your video files (AVI/MP4) in the dataset/videos/ directory.
2. Run the script:

   python src/video_dataset_prep.py
   

The script will scan all video files, validate them, and display their properties (filename, FPS, total frames, duration).

Output for STEP-1

For each valid video:

• Video filename
• FPS
• Total frames
• Duration in seconds

For invalid videos, an error message is logged, and processing continues.

STEP-2: Frame-by-frame Video Traversal and Timestamp Mapping

This step performs frame-by-frame traversal of videos and calculates timestamps for each frame.

Usage for STEP-2

1. Ensure videos are in dataset/videos/ directory.
2. Run the script:

   python src/video_frame_traversal.py
   

The script will process each video frame-by-frame, calculate timestamps, and log progress every 100 frames.

Output for STEP-2

For each video:

• Processing start message with FPS
• Progress logs: Frame number, timestamp in HH:MM:SS and seconds
• Completion message with total frames processed

STEP-3: Face Detection from Video Frames

This step performs face detection on each frame using Haar Cascade.

Prerequisites for STEP-3

1. Install MTCNN:

   pip install mtcnn
   

   Or ensure it's in requirements.txt.

Usage for STEP-3

1. Ensure videos are in dataset/videos/ and Haar Cascade XML is in models/.
2. Run the script:

   python src/face_detection.py
   

The script will process each video frame-by-frame, detect faces using MTCNN, and save cropped faces.

Output for STEP-3

When faces are detected and saved:

• Video name, frame number, timestamp, number of faces saved

Faces are saved in dataset/faces/<video_name>/face_f<frame>_t<timestamp>_p<index>.jpg

STEP-4: Face Cropping and Intelligent Saving

This step crops detected faces, resizes them, and saves them intelligently to avoid duplicates.

Usage for STEP-4

1. Ensure videos are in dataset/videos/ and Haar Cascade XML is in models/.
2. Run the script:

   python src/face_cropping.py
   

The script will process each video, detect faces, crop and resize them (160x160), and save all faces at fixed intervals (every 30 frames).

Output for STEP-4

When faces are saved (every 30 frames):

• Video name, frame number, timestamp, number of faces saved

Faces are saved in dataset/faces/<video_name>/face_f<frame>_t<timestamp>_p<index>.jpg

Project Structure

• dataset/videos/: Place your video files here (AVI and MP4 formats)
• dataset/faces/: Cropped face images saved here, organized by video
• src/: Source code
• models/: DNN model files (prototxt, caffemodel) and other models
• logs/: Reserved for future logging (not used in current steps)

Requirements

• Python 3.11+
• OpenCV

Installation

1. Clone or download the project.
2. Install dependencies:

   pip install -r requirements.txt
   

Troubleshooting

• Ensure videos are in dataset/videos/ directory.
• Supported formats: AVI, MP4.
• If OpenCV fails to read a video, check if the file is corrupted or in an unsupported codec.