Human-Activity-Recognition

This project implements a real-time human activity recognition system using deep learning models. The project integrates YOLO for object detection and ResNet-34 for action classification, offering a robust solution for applications like surveillance, healthcare monitoring, and more.

---

Features

• Real-time person detection using YOLO (You Only Look Once).
• Activity classification using a pre-trained ResNet-34 model.
• Supports both image-based and video-based human activity recognition.
• GPU-accelerated performance with CUDA for faster inference.
• Audio feedback for detected actions and status.

---

Folder Structure

model/
├── resnet-34.onnx                      
├── yolov3.cfg                          
├── yolov3.weights                      
├── coco.names                          
└── Audio/                              
images/
test/                                
train/
├── class 1                      
├── class 2                           
image.py                                
video_detection.py                      
realtime.py                             

---

Installation

Prerequisites

• Python 3.8 or later
• Libraries:
  • opencv-python
  • numpy
  • playsound
  • CUDA-compatible GPU and drivers for acceleration

Setup

1. Clone the repository:

   git clone https://github.com/harish-00-11/Human-Activity-Recognition-using-deep_learning.git
   cd real-time-human-activity-recognition

2. Install required libraries:

   pip install -r requirements.txt

3. Download the required pre-trained models:

   • Place the YOLO files (yolov3.cfg, yolov3.weights, and coco.names) in the model/ directory.
   • Rename resnet-34_kinetics.onnx to resnet-34.onnx and place it in the same directory.

4. Add your test images to the images/ folder and videos to the test/ folder.

---

Usage

1. Image-Based Recognition

Run the following command to classify actions from images:

python image.py

2. Video-Based Recognition

Process a video file to detect and classify actions:

python video_detection.py

3. Real-Time Recognition

Use a webcam to detect and classify actions in real-time:

python realtime.py

---

Outputs

• Bounding Boxes: Persons detected by YOLO are highlighted with green rectangles.
• Action Labels: The identified activity is displayed on the screen with confidence percentages.
• Audio Feedback: Alerts indicating detection status and classified actions.

---

---

Acknowledgments

• YOLOv3: For real-time object detection.
• ResNet-34: For robust activity recognition.
• OpenCV: For image processing and display.
• CUDA: For accelerating the computation.

---