Vehicle Speed Estimation

This project estimates the speed of vehicles in a video using YOLOv10 for object detection and DeepSORT for object tracking. It calculates speed based on pixel distance and perspective transformation, overlaying the results and bounding boxes on the output video.

Demo

Features

• Object Detection: Uses yolov10n.pt for fast and accurate detection.
• Object Tracking: Implements DeepSORT to track vehicles across frames.
• Speed Estimation: Calculates speed in km/h using perspective transformation.
• Blurring: Option to blur specific objects (e.g., license plates or specific vehicle classes) for privacy.

Installation

1. Clone the Repository

git clone https://github.com/lovnishverma/vehicle-speed-estimation.git
cd vehicle-speed-estimation

2. Set up the Environment

It is recommended to use a virtual environment to manage dependencies. (conda is optional)

Windows:

python -m venv venv
venv\Scripts\activate

macOS / Linux:

python3 -m venv venv
source venv/bin/activate

3. Install Dependencies

pip install -r requirements.txt

4. Prepare Input Files (i have already placed in content folder you can skip this part)

The script requires an input video. You can download the sample videos used in development or place your own video in a content/ folder.

mkdir content
# Download sample highway footage
wget -P content https://github.com/AarohiSingla/Speed-detection-of-vehicles/raw/main/highway.mp4

Note: The yolov10n.pt model weights will be downloaded automatically by the Ultralytics library on the first run.

Usage

Run the main script object_tracking.py with the following arguments:

Basic Run

Run with default settings (processes content/highway.mp4):

python object_tracking.py

Custom Arguments

You can specify input/output paths, confidence thresholds, and specific classes to track or blur.

Argument	Description	Default
--video	Path to the input video file.	content/highway.mp4
--output	Path to save the processed video.	content/output.mp4
--conf	Confidence threshold for detection.	0.5
--class_id	Specific class ID to track (e.g., 2 for cars).	None (Tracks all)
--blur_id	Class ID to apply Gaussian blur to.	None

Examples

Process a custom video:

python object_tracking.py --video data/my_traffic.mp4 --output results/final.mp4

Track only cars (COCO Class ID 2):

python object_tracking.py --class_id 2

Blur trucks (COCO Class ID 7) while tracking:

python object_tracking.py --blur_id 7

Configuration

Perspective Transform

To get accurate speed estimations for different camera angles, you must calibrate the perspective transform points in object_tracking.py.

Modify the SOURCE_POLYGONE array to match the road layout in your specific video:

# object_tracking.py
SOURCE_POLYGONE = np.array([
    [x1, y1], # Top-left
    [x2, y2], # Top-right
    [x3, y3], # Bottom-right
    [x4, y4]  # Bottom-left
], dtype=np.float32)

Acknowledgements

• YOLOv10: Ultralytics
• DeepSORT: Simple Online and Realtime Tracking