DeepHPD for Smart homes

DeepHPD is a novel application which uses Deep Learning in order to detect the presence of a human being. This is done by using a combination of Convolutional Neural Networks and OpenCV which along with a Raspberry pi and Basler Dart Camera, which blinks an LED when it detects human presence.


The project was deployed on hardware stack which consisted of  Raspberry Pi3 running Raspbian 8.0 Jessie and the streaming was done using Basler Dart Camera model daA2500-14uc.

The Software stack includes:

  1. Tensorflow

    1. CUDA
    2. Numpy
    3. OpenCV
    4. RPi GPIO

The Convolutional neural network used Transfer Learning and was trained using the weights of the InceptionV3 model and used CUDA for GPU based training and achieved a classification accuracy of 96%.

In order to use the components of the project

  1. Ensure Python 3 is installed and all software dependencies are installed.




pip install opencv-python


sudo apt-get install RPi-GPIO

  1. Following this you should download the folder from this link for the weights of the trained model from link provided in the GitHub repository

  2. Following this, run python script, and the classification is generated following which the labels are passed to the Raspberry Pi via RPi-GPI

The camera captures the image and sends the status via labels to the Raspberry Pi3 as an input which is linked to the blink LED program turning it on and off depending on the presence or absence of a human.


Led ON in presence of human LED OFF when no human detected. OUTPUT


GitHub Repository


Sree Harsha Nelaturu
pushed 61587d32dc5b479009f33a70775b8954b104bde1
2017-11-25 10:51:52 UTC
Sree Harsha Nelaturu
pushed 33fb51da85e180707d15f60ba13cbb90eb0e2aec
2017-11-25 10:48:24 UTC
Sree Harsha Nelaturu
pushed a20c357084b28bf69241ead2d4c0fe84d76b751f
2017-11-25 10:48:06 UTC
Sree Harsha Nelaturu
pushed 1eaee1487d8e0dc9ec38fb8a089aa16205be1cfe
2017-11-25 10:47:22 UTC
Sree Harsha Nelaturu
pushed 5c15b19fa26c4ad8c237ea4b575d2d4fc93f61f6
2017-11-25 10:43:32 UTC
Sree Harsha Nelaturu
pushed 202d81828938e5cb140a6e2ccc975a182ce4e048
2017-11-25 07:28:32 UTC
Sree Harsha Nelaturu
pushed 69938318ae9360239d5d895df752bfb0243659e9
2017-11-24 18:17:35 UTC
Sree Harsha Nelaturu
pushed d661d20dc2792fe0d16af7bba504d2f4efab00e2
2017-11-24 18:16:45 UTC
Sree Harsha Nelaturu
pushed 97f5a2e65a941d698d337c24e4f490eae66e8228
2017-11-24 18:15:12 UTC
Sree Harsha Nelaturu
pushed c4e2c2407d285cef0538bfd18906eebfdc3c272b
2017-11-24 18:08:25 UTC
Sree Harsha Nelaturu
pushed dae6d13065a524afb7e65b76bcbd8c97f454b4a8
2017-11-24 18:08:08 UTC
Sree Harsha Nelaturu
pushed 183802846488b0d7b259ef47dc0e03f90407cbf7
2017-11-24 18:07:41 UTC
Sree Harsha Nelaturu
pushed 0469e29fb5efa6992a0a780a3c755764bddf16b2
Add files via upload
2017-11-24 18:06:02 UTC
Sree Harsha Nelaturu
pushed b6591c424c1b5ae952390a293a4b936de98c547c
Initial commit
2017-11-24 17:51:00 UTC
Title Description Format
Schematics for Pi png
Bill of materials
Name Article number Link Quantity Unit Price
Raspberry Pi 3 KIT 1 50.00 $
LED 1 0.10 $
Jumper Wires 15 2.00 $
Basler DART 1
Total 80.10 $

Project State

Public Project


Software Licence: MIT
Hardware Licence: Cern OHL 1.2

Project Tags




Does this project pique your interest?

Login or register to join or follow this project.

Himanshu_S report abuse
The whole project can be build just using a micro controller and a PIR is this project relevant for smart homes?
im_anithp report abuse
@Shanque thank you for your comment, but when you compare the project with a basic electronic project then you are wrong, this an advanced AI project where we have trained the model using CNN, thus can detect things in more stochastic manner. And apart from this it also detects the human face, and as the project is concerned i have made a project that i mentioned in the pitch. Best, Anith
Himanshu_S report abuse
You have used basler dart camera? How u clicked image with it......can u send me the code u used? git hub code is not working.
im_anithp report abuse
I have used openCV for image processing. And in the code itself using cuda using deploy it on your GPU. @Shanque
anirban-deb report abuse
Which framework are you using to train - squeezenet or Keras ? And have you managed to train it to distinguish between regular house occupants vs someone else ? ( that would be helpful in a security camera )
Back to top

Your comments, please!

Want to comment this ... Show more