Trajectory Forecast Multi Object Tracking

Trajectory Forecast is a lightweight, modular extension built on top of Ultralytics YOLO that enables real-time multi-object tracking with future motion prediction. It combines detection, tracking, motion history modeling, and velocity-based forecasting into a unified pipeline that can be used both as a command-line tool and as a Python library. The system is designed for practical computer vision applications such as traffic analytics, surveillance systems, robotics pipelines, and edge AI deployments. Unlike heavy deep-learning forecasting architectures, this framework uses a lightweight motion model optimized for stability and speed, making it suitable for:

• Autonomous systems
• Intelligent surveillance
• Traffic analytics
• Robotics perception pipelines
• Edge AI deployments

Demo

trajectory-forecasting-demo-github.mp4

The demo shows real-time object tracking with projected future trajectories, where predicted points indicate the expected motion path.

Key Features

Real-Time Performance

Designed for low-latency environments such as robotics and surveillance.

Multi-Object Tracking

Supports advanced trackers including:

• ByteTrack
• BoT-SORT

Motion Forecasting

Predicts future object trajectories using velocity-based projection.

Modular Design

The framework is structured as reusable modules that can be integrated into larger vision systems.

CLI + Python API

Use as:

• Command-line application
• Python library

System Architecture

Video Input
     │
     ▼
YOLO Object Detection
     │
     ▼
Multi-Object Tracker
(ByteTrack / BoT-SORT)
     │
     ▼
Trajectory History Buffer
     │
     ▼
Velocity Estimation
     │
     ▼
Trajectory Forecast Module
     │
     ▼
Future Motion Visualization

Installation

Clone the repository:

git clone https://github.com/arafathosense/trajectory-forecast.git
cd trajectory-forecast

Install dependencies:

pip install -r requirements.txt

Recommended environment:

Python ≥ 3.10
CUDA-enabled GPU (optional)

Configuration

Create a configuration file named config.yaml.

conf: 0.5
tracker: "bytetrack.yaml"
classes: [2, 3, 5]
history: 40
min_points: 8
forecast_steps: 35
vel_window: 10
ema_alpha: 0.7
forecast_color: [255, 0, 0]

Parameter Description

Parameter	Description
conf	Detection confidence threshold
tracker	Tracker configuration file
history	Number of frames used for trajectory history
forecast_steps	Number of future trajectory points
vel_window	Frames used for velocity estimation
ema_alpha	Smoothing coefficient

CLI Inference

trajectory-forecast \
--model yolo26n.pt \
--source "https://tinyurl.com/bddswzba" \
--config "path/to/config.yaml"

Supported input sources:

• Webcam streams
• Video files
• Network streams
• Image sequences

Python API

The framework can be embedded into Python pipelines.

from tf import run_inference
from tf.config import ForecastConfig

config = ForecastConfig(
    conf=0.5,
    forecast_steps=50,
    ema_alpha=0.7,
    classes=[0, 2, 5, 6, 7]
)

run_inference(
    model="yolo26n.pt",
    source="video.mp4",
    config=config
)

Forecasting Methodology

The trajectory forecasting module is based on lightweight motion modeling.

Pipeline components include:

1. Object Center Extraction

Bounding box centers are extracted from tracked objects.

2. Exponential Moving Average

Noise reduction using trajectory smoothing.

3. Sliding Window Velocity Estimation

Velocity is estimated using median displacement over recent frames.

4. Linear Motion Projection

Future positions are predicted using velocity extrapolation.

5. Stationary Filtering

Objects with near-zero movement are filtered to prevent unstable predictions.

This design provides stable real-time trajectory prediction without requiring computationally expensive deep neural networks.

Example Applications

Trajectory prediction is useful in many AI domains:

Application	Example
Autonomous Driving	Vehicle path prediction
Traffic Monitoring	Flow analysis
Robotics	Motion planning
Security Systems	Suspicious movement detection
Smart Cities	Crowd analytics

Performance

Example benchmark (RTX GPU):

Model	FPS	Forecast Steps
YOLO-Nano	~120 FPS	30
YOLO-Small	~85 FPS	35
YOLO-Medium	~55 FPS	40

Actual performance depends on GPU, video resolution, and tracker configuration.

Acknowledgments

This project builds upon the open-source computer vision ecosystem developed by:

• Ultralytics for YOLO detection models
• Open-source multi-object tracking research community

Contributing

Contributions are welcome.

You can contribute by:

• Improving trajectory prediction algorithms
• Adding new trackers
• Improving performance optimization
• Enhancing documentation

Please open an issue or submit a pull request.

License

Released under the MIT License.

👤 Author

HOSEN ARAFAT

Software Engineer, China

GitHub: https://github.com/arafathosense

Researcher: Artificial Intelligence, Image Computing, Image Processing, Machine Learning, Deep Learning, Computer Vision