AI Ganpati

Contents

• Introduction
• Repository contents
• Flow Diagram
• What goes in?
• What comes out?
• Dependencies
• For Embedded Linux platforms
• Docker Image

---

Introduction

• This repository contains a fun application made on python3 intended to dispense sweets to individuals based on their choice.

• Before running the codes, we need to install all the prerequisites.

• Have made a requirements.txt for the same.

• Use pip3 install -r requirements.txt to install prerequisites.

• Can be executed on x86 and ARM architecture systems.

• Have also created a Dockerfile and docker image for the application.

---

Repository contents

1. requirements.txt - contains all the prerequisites required for the functioning of the application.

2. models - conatins trained models required for face inference.

   Note: The voice model has been pre-trained using several voice samples.

3. utils - contains the all dependencies that the main codes require.

4. words - contains the voice input given by user which will be used to get the label.

5. scripts - contains script files to perform various actions using Arduino GPIO pins.

6. arduino-codes - codes for some action to be performed.

7. Dockerfile - dockerfile for the application

---

Block Diagram

***

Input

• Run python3 input.py for face input and then voice input for choice of sweet (laddu/modak/pedha).
• LEDs can be used to display the status to a user. (see ./Scripts)
• The faces in the frame are first detect, aligned and croppped.
• Then the features of the face are extracted and converted to 128D embeddings.
• The user says the speech of their choice, which is saved to a .wav format and given to the speech model.
• The label from the audio file is saved along with the embeddings of the person in the database.

---

Output

• Run python3 recognition.py
• When a face is detected, the features of the face are extracted.
• The system verifies if the person exists in the database.
• On recognition, the corresponding label is displayed (person name/unknown)
• Dispenses the sweet as per the registered person's choice.
• Both the codes (input.py and recognition.py) can be executed simultaneously.

---

Dependencies

All the required dependencies can be install by running the command pip install -r requirements.txt

---

For Embedded Linux Platforms

• Connect 2 USB/CSI cameras (as per availability) and a mic for voice input
• Install drivers if required using insmod [driver location]
• Check the device IDs for the connected peripherals in /proc/asound/
• Make changes in .asoundrc accordingly.

---

Docker Image

• Install docker using sudo apt-get install docker.io and assign sudo permission to it.

• You can find the readymade image that I've already built using docker pull darpanjain/ai-input

• Visit My DockerHub Profile

• Run the image using docker run -it --rm ai-input

• You can use the provided Dockerfile to build your own image.

1. Clone the repo to your system
2. Build your image using sudo docker build -t application:v1 .

---