shot-detector-cnn : Video Processing with a CNN Model

Description

This project implements a video processing pipeline using a Convolutional Neural Network (CNN) model. The goal is to extract images from a video, preprocess them, and use them to train a classification model.

Project Structure

The project is organized as follows:

• projet.mp4 : Source video
• extracted_frames/ : Folder containing extracted images
• tp_VIDEO_CNN/ : Contains scripts and resources for the CNN
• CONSIGNES.pdf : Project instructions
• annotated_video.mp4 : Annotated video after analysis
• video_test_3.mp4 : Test video

Main Features

1. Frame Extraction:

   • Load the video
   • Convert frames to grayscale
   • Store frames in a folder

2. Image Preprocessing:

   • Classify frames into two categories: shoot and no_shoot
   • Convert images into numpy arrays
   • Normalize and resize images

3. CNN Model Training:

   • Build a CNN model using TensorFlow/Keras
   • Train the model on the extracted images
   • Validate the model

4. Evaluation and Visualization:

   • Display model performance
   • Visualize accuracy and loss curves
   • Apply the model to a test video

Execution

Prerequisites

Make sure you have installed the following dependencies:

pip install numpy tensorflow opencv-python scikit-learn matplotlib

Run Frame Extraction

python extract_frames.py

Train the Model

python train_model.py

Test the Model

python test_model.py

Results

• The model achieves an accuracy of 97-100% on the validation set.
• The accuracy and loss curves show good convergence.
• The model accurately identifies shooting scenes.

Possible Improvements

• Optimize the model architecture
• Increase data augmentation to improve robustness
• Integrate a real-time detection system

Author

Andre KHELLA

---

License: This project is open-source. You are free to modify and share it while respecting copyright rules.