🧠 Object Detection using OpenCV and a Pretrained Model

This project demonstrates real-time object detection using OpenCV and a pre-trained SSD MobileNet v3 model trained on the COCO dataset.

🚀 How It Works

The project uses the TensorFlow Object Detection API along with OpenCV's dnn module to perform object detection. Here’s how the system works:

1. Model and Config Files

   • frozen_inference_graph.pb: The pre-trained TensorFlow model containing the learned weights.
   • ssd_mobilenet_v3_large_coco_2020_01_14.pbtxt: Configuration file that defines the model architecture for OpenCV.
   • coco.names: A text file that lists all 80 class names from the COCO dataset used for labeling detections.

2. Preprocessing Input

   • Input frames are converted into blobs using cv2.dnn.blobFromImage() before being passed to the neural network.

3. Detection Logic

   • The model returns bounding boxes, class IDs, and confidence scores.
   • Detections with confidence above a set threshold (e.g., 0.5) are visualized using OpenCV.

📁 Project Structure

File/Folder	Description
main.py	Main script that loads the model, performs detection, and visualizes results using OpenCV
coco.names	List of class labels used by the COCO-trained model
frozen_inference_graph.pb	Pre-trained model (frozen graph) used for inference
ssd_mobilenet_v3_large_coco_2020_01_14.pbtxt	Configuration file for the SSD MobileNet v3 model
lena.png	Sample image for testing object detection
venv/	Python virtual environment with required packages installed
README.md	Project documentation (this file)

🛠️ Requirements

Make sure to install the required packages:

pip install opencv-python

If using the provided virtual environment, activate it and install the packages from requirements.txt:

source venv/bin/activate  # On Windows: venv\Scripts\activate
pip install -r requirements.txt

▶️ Running the Project

Run the object detection script with:

python main.py

You can modify the script to use a webcam feed or replace lena.png with another image for testing.

📦 Model Info

• Model: SSD MobileNet v3
• Trained On: COCO dataset (80 classes)
• Framework: TensorFlow
• Inference: Performed using OpenCV’s cv2.dnn module for efficiency and portability

📽️ Demo

Watch the object detection in action: YouTube Demo