Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Multi-camera live traffic and object counting with YOLO v4, Deep SORT, and Flask.
| Date | Stars |
|---|---|
| 2026-07-24 | 1107 |
| 2026-07-25 | 1107 |
| 2026-07-28 | 1107 |
| 2026-07-30 | 1107 |
| 2026-08-06 | 1108 |
Today
+1 stars today
This week
+1 stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.09%/day
<h1 align='center'>
Multi-Camera Live Object Tracking
</h1>
This repository contains my object detection and tracking projects. All of these can be hosted on a cloud server.
You can also use your own IP cameras with asynchronous processing thanks to [ImageZMQ](https://github.com/jeffbass/imagezmq). I've written a blog post on how to stream using your own smartphones with ImageZMQ [here](https://leonlok.co.uk/blog/live-video-streaming-using-multiple-smartphones-with-imagezmq/).
## Deep SORT and YOLO v4
Check out my [Deep SORT repository](https://github.com/LeonLok/Deep-SORT-YOLOv4) to see the tracking algorithm that I used which includes the options for Tensorflow 2.0, asynchronous video processing, and low confidence track filtering.
***
## Traffic Counting ([Link](https://github.com/LeonLok/Multi-Camera-Live-Object-Tracking/tree/master/traffic_counting))
This project is an extension of the object counting app.
<div align='center'>
<img src="gifs/traffic_counting1.gif" width="80%"/>
</div>
### ([Full video](https://www.youtube.com/watch?v=x6vkXf-mgaw&feature=youtu.be))
### Features
* Trained using a total of **244,617** images generated from the DETRAC dataset. You can find the conversion code that I created [here](https://github.com/LeonLok/Multi-Camera-Live-Object-Tracking/tree/master/detrac_tools).
* I used this [paper](https://ieeexplore.ieee.org/document/8909903) as a guideline for data preparation and training.
* Only counts each tracking ID once.
* Counts objects by looking at the intersection of the path of the tracked object and the counting line.
* Hence, those that lose tracking but are retracked with the same ID still get counted.
* Tracked using low confidence track filtering from the same [paper](https://ieeexplore.ieee.org/document/8909903).
* Offers much lower false positive rate.
* Tracked objects show average detection confidence.
* Tracked classes determined by most common detection class.
* Showing detections is optional (but hides average detection confidence).
* Multiple IP cameras possible.
* Video streaming possible via emulated IP camera.
* Directional counts can be configured based on angle.
* Records counts for every set interval of the hour.
* Total count.
* Class-based counts.
* Records intersection details for each counted object.
* Time of intersection.
* Coordinate of intersection.
* Angle of intersection.
* Can be hosted on a cloud server.
Note that since DETRAC doesn't contain any motorcycles, they are the only vehicles that are ignored. Additionally, the DETRAC dataset only contains images of traffic in **China**, so it struggles to correctly detect certain vehicles in other countries due to lack of training data. For example, it can frequently misclassify hatchbacks as SUVs, or not being able to detect taxis due to different colour schemes.
***
## Object Counting ([Link](https://github.com/LeonLok/Multi-Camera-Live-Object-Tracking/tree/master/object_counting))
This project was originally intended to be an app for counting the current number of people in multiple rooms using my own smartphones, where the server would be remotely hosted. Below shows detection, tracking, and counting of people and cars.
<div align='center'>
<img src="gifs/object_counting2.gif" width="50%"/>
</div>
### Features
* Counts the current number of objects in view.
* Tracking is optional.
* Multiple IP cameras possible.
* Records current counts for every set interval of the hour.
* Current total count.
* Current class-based counts.
* Can be hosted on a cloud server.
***
## Using my own smartphones as IP cameras
<div align='center'>
<img src="gifs/object_counting1.gif" width="50%"/>
</div>
***
## Training your own vehicle tracking model ([Link](https://github.com/LeonLok/Multi-Camera-Live-Object-Tracking/tree/master/detrac_tools))
I trained a YOLO v4 and Deep SORT model using the [DETRAC](http://detrac-db.rit.albany.edu/) training dataset with v3 annotatiExcerpt of 7,640 characters
Read on GitHub
71
3
1
Would you bet a product on this? Bounded 0–100 and slow moving.
matched fp:f6ccc981b7064d33, topic:object-detection, topic:yolo, topic:opencv