Human Detection using Deep Learning

Project Overview

This project focuses on building and evaluating deep learning models for human detection using image data. The models implemented include a custom CNN (Convolutional Neural Network), VGG16, Xception, and ResNet50. The project involves data loading and preprocessing, model training, evaluation, and visualization of the results.

Dataset

The dataset used in this project is the Human Detection Dataset, which can be downloaded from Kaggle.

Requirements

• Python 3.6+
• Jupyter Notebook or Google Colab
• Libraries: numpy, pandas, matplotlib, seaborn, sklearn, tensorflow, keras, opencv-python

Project Structure

.
├── data/
│   ├── human-detection-dataset/
│   └── ...
├── notebooks/
│   ├── data_preprocessing.ipynb
│   ├── cnn_model.ipynb
│   ├── vgg16_model.ipynb
│   ├── xception_model.ipynb
│   └── resnet_model.ipynb
├── models/
│   ├── cnn_model.h5
│   ├── vgg16_model.h5
│   ├── xception_model.h5
│   └── resnet_model.h5
└── README.md

Setup and Usage

Step 1: Clone the Repository

git clone https://github.com/yourusername/human-detection.git
cd human-detection

Step 2: Install Dependencies

pip install -r requirements.txt

Step 3: Download and Prepare the Dataset

Download the dataset from Kaggle and place it in the data/ directory.

Step 4: Data Preprocessing

Run the data_preprocessing.ipynb notebook to load, preprocess, and split the dataset into training and testing sets.

Step 5: Train the Models

You can train different models by running the corresponding notebooks:

• cnn_model.ipynb
• vgg16_model.ipynb
• xception_model.ipynb
• resnet_model.ipynb

Step 6: Evaluate the Models

Each notebook includes code to evaluate the models using metrics such as accuracy, F1 score, ROC AUC, and confusion matrix. The evaluation results are visualized using plots.

Results

The results include:

• Training and validation loss and accuracy plots
• ROC AUC curves
• Confusion matrices
• Classification reports
• Accuracy and F1 scores

Model Comparison

Model	Accuracy	F1 Score	ROC AUC
CNN	62.34%	74.34	0.68
VGG16	83.98%	86.93	0.92
Xception	85.28%	87.31	0.94
ResNet50	71.43%	79.11	0.82

Conclusion

This project demonstrates the application of various deep learning models for human detection. It provides a comprehensive comparison of different models' performance and highlights the effectiveness of transfer learning using pre-trained models like VGG16, Xception, and ResNet50.

Future Work

• Experiment with other architectures like InceptionV3, DenseNet, etc.
• Perform hyperparameter tuning for better model performance.
• Deploy the best-performing model as a web service.

References

• Kaggle Human Detection Dataset
• TensorFlow Keras Applications