Object Dimensions Detector using a Single Camera

This project demonstrates a Streamlit application for estimating the real-world dimensions (width, height, depth) and distance of objects from standard 2D images using a single camera (or dual front/side views). It leverages a pre-trained monocular depth estimation model and a user-defined reference object for scale calibration.

Overview

The application takes one or two images (front and side views), camera parameters (focal length or field of view), and the known dimension of a reference object as input. It then:

1. Estimates Depth: Uses a pre-trained DPT model to generate a relative depth map for each input image.
2. Calibrates Scale: Allows the user to draw a bounding box around a reference object with a known real-world dimension to establish a metric scale.
3. Detects Objects: Automatically detects potential target objects in the image based on depth discontinuities or proximity.
4. Estimates Dimensions: Calculates the width, height, and distance of the detected objects based on their pixel size, estimated depth, and the calibrated scale.
5. Combines Views (Optional): If two views are provided, it combines the width and height from the front view with the depth and height from the side view to provide a 3D dimension estimate.

Screenshots

Input and Depth Estimation (Front View)

Example of an uploaded front view image.

The relative depth map estimated by the DPT model for the front view.

Object Detection

Visualization of the object detection process based on the depth map.

Measurement Results (Front View)

The front view image with detected objects and their estimated dimensions overlaid.

Example Combined 3D Results

The application displays the estimated dimensions numerically. For example:

Dimension	Estimated Value
Width (from Front)	8.5 cm
Depth (from Side)	8.5 cm
Height (Avg)	15.5 cm
Distance (Front)	0.20 m

Key Features

• Single/Dual Image Input: Supports measurement from a single front view or a combination of front and side views for 3D estimation.
• Reference Object Calibration: User-friendly interface to select a reference object and input its known dimension.
• Camera Parameter Flexibility: Accepts either focal length (in pixels) or horizontal/vertical field of view (in degrees).
• Automatic Object Detection: Attempts to automatically identify and measure multiple objects in the scene based on depth information.
• Streamlit Interface: Provides an intuitive web-based interface for easy interaction.
• Visualization: Overlays bounding boxes and estimated dimensions on the input images.

Methodology

The core steps involved are:

1. User Input: Uploading images, specifying reference object dimensions, and providing camera parameters.
2. Depth Prediction: Utilizing the Intel/dpt-hybrid-midas pre-trained model from Hugging Face Transformers to predict depth maps.
3. Scale Calibration: Calculating a scale factor based on the reference object's known size, its pixel size in the image, its relative depth, and the camera parameters.
4. Object Detection: Employing image processing techniques on the depth map (edge detection, thresholding, connected components analysis) to identify potential objects.
5. Dimension Estimation: Using the calibrated scale, pixel dimensions, and estimated depth to calculate the real-world width, height, and distance of the detected objects.
6. Two-View Fusion: Combining measurements from front and side views (if available) to estimate the 3D dimensions.

Limitations

• Reliance on Depth Model Accuracy: The accuracy of the results is heavily dependent on the quality of the depth maps generated by the pre-trained model.
• Camera Parameter Sensitivity: Incorrect focal length or field of view values will lead to inaccurate estimations.
• Reference Object Requirement: Accurate selection and known dimensions of a reference object are crucial.
• Object Detection Challenges: Depth-based object detection can be unreliable in complex scenes.
• Two-View Assumptions: The 3D combination assumes specific object orientations and alignment between views.

Getting Started with the Object Dimensions Detector

This guide will walk you through the steps to get the Object Dimensions Detector application running.

Prerequisites

Before you begin, ensure you have the following installed on your system:

• Python 3.9: The application is written in Python.
• pip: Python package installer (usually comes with Python).
• Git (optional but recommended): For cloning the repository.

Installation

1. Clone the Repository (Optional):

   You can clone the Git repository using:

   git clone https://github.com/HarshdeepJ/Object_Volume_Detector.git
   cd Object_Volume_Detector

2. Install Dependencies:

   The application relies on several Python libraries. You can install them using pip. Make sure you are in the project directory (where the requirements.txt file, if provided, is located)

   pip install -r requirements.txt

Running the Application

1. Navigate to the Project Directory:

   Open your terminal or command prompt and navigate to the directory where app.py is located.

2. Run the Streamlit Application:

   Use the streamlit run command followed by the name of the main Python script:

   streamlit run app.py

   Replace app.py with the actual name of the Python file that contains the Streamlit application code.

Using the Application

1. Upload Images: Follow the instructions in the application to upload one (front view) or two (front and side view) images.
2. Specify Reference Object: Enter the known real-world width of a reference object present in your image(s) (e.g., the width of an ID card).
3. Provide Camera Parameters: Enter either the focal length of your camera in pixels or the horizontal and vertical field of view (FoV) in degrees. Accurate parameters are crucial for accurate measurements. You might need to look up these specifications for your camera (e.g., smartphone camera).
4. Select Reference Object: Use the interactive drawing tool in the application to draw a bounding box around the reference object in the uploaded image.
5. Run Measurement: Click the button or follow the prompts in the application to initiate the depth estimation and dimension measurement process.
6. View Results: The application will display the detected objects (if any) with their estimated dimensions and distances overlaid on the image(s), along with numerical results.
7. Two-View Analysis (Optional): If you uploaded a second (side view) image, follow the same steps for that image to obtain depth and height measurements, which will then be combined with the front view results.