ndi_facetrack's Introduction

NDI_FaceTrack

A Windows desktop application allows Bird Dog PTZ Cameras to automatically track persons of interests, freeing up camera operators. This application uses Machine Learning and CV Techniques to Identify and track both faces and human body figures.

Quick Start

Highly suggest to use a virtual environment when running and install the required modules

python -m virtualenv .venv
pip install -r requirements.txt'

Launch the FastAPI server. By defaults it runs on 127.0.0.1:8000 Run the Tracking GUI as the direct interface to the program

python TrackingServer_FastAPI.py
python TrackingGUI.py

Packaging

This project supports packaging the program to an .exe using PyInstaller cli.spec is given for easier packaging

Architecture

Backend: The backend framework is built from a pure FastAPI framework. The FastAPI handles communication between external applications using websockets and to start namedPipes(FIFO) between the Tracking Module and the GUI. Currently it supports data exchanged between:

Websockets via Python Pickling
Websockets via Qt Framework
namedPipes via Python Pickling

Frontend: The GUI interface available in TrackingGUI.py is built on top of the PySide2 (Qt) framework. It communicates with the FastAPI backend using namedPipes(FIFO).

NDI Facetracking uses open multiple open source projects to work seamlessly betwen each other, namely:

A self-implemented state-of-the-art tracking algorithm based on the following paper: https://www.hindawi.com/journals/mpe/2019/
A robust and fast facial detection method using MTCNN: https://github.com/ipazc/
OpenCV Object Trackers
Custom Tiny Yolov4 running on OpenCV DNN
NDI C++ / Python Wrapper Library: https://github.com/buresu/ndi-python

https://github.com/Tomas1337/NDI_FaceTrack

An integral part of this project is to have this easily deployable to users in a robust lightweight executable file. Currently achieves +30FPS on an QuadCore Intel 2.4GHz

ndi_facetrack's People

Contributors

Stargazers

Watchers

ndi_facetrack's Issues

Tracking on the 'edge' becomes problamtic

We have to make sure that the x,y,w,h variables upon on track_with_csrt return and setting of any class variables related to it needs to be managed.

Where in the x,y,w,h, can never reach the extremes of the frame. Even if the tracked object is at the edge, the x needs to be a certain distance from the edge and the w needs to have padding from the edge as well. Basically the bounding box can never touch the extremes of the frame with an padding of about 100 pixels.

Instead of using .jpg encode to send through pipe, use pure numpy instead

Need to change this to numpy and send through pipe. Must also handle the decoding of the pipe.
Better: Develop a feature where we can switch from each.

        #image_payload = PipeClient_Image_Payload(frame = jpeg.encode(frame))
        is_success, im_buf_arr = cv2.imencode(".jpg", frame)
        byte_im = im_buf_arr.tobytes()
        
        image_payload = PipeClient_Image_Payload(frame = byte_im)
        self.reset_trigger = False
        self.pipeClient.writeToPipe(payload = image_payload.pickle_object())

Solves: LAG

Recommend Projects

tomas1337 / ndi_facetrack Goto Github PK

ndi_facetrack's Introduction

NDI_FaceTrack

Quick Start

Packaging

Architecture

ndi_facetrack's People

Contributors

Stargazers

Watchers

Forkers

ndi_facetrack's Issues

Tracking on the 'edge' becomes problamtic

Instead of using .jpg encode to send through pipe, use pure numpy instead

Recommend Projects

React

Vue.js

Typescript

TensorFlow

Django

Laravel

D3

Recommend Topics

javascript

web

server

Machine learning

Visualization

Game

Recommend Org

Facebook

Microsoft

Google

Alibaba

D3

Tencent