Object-Detection-with-Voice-Feedback-using-Raspberry-Pi-and-Tensorflow-Lite

Introduction

In this Project, I teach you how to set up the TensorFlow Lite and Voice Feedback on the Raspberry Pi and use it to run object detection models with voice feedback. In this work, I will use the Raspberry Pi 3 or Raspberry Pi 4 running either Raspbian Buster or Raspbian Stretch.(Raspbian OS Repo Link: https://downloads.raspberrypi.org/raspbian/images/) try Raspbian images(2018-2020), They are the supported OS images.

Lesson 1 - How to Set Up TensorFlow Lite on the Raspberry Pi

Step 1. Check Python Version in Raspberry Pi

In this tutorial Tensorflow Lite installation currently supported python version is 3.5, 3.6, 3.7 & 3.8 (3.9 sometime came package or library missing issue)..

python3 --version

Step 2. Update the Raspberry Pi

sudo apt update

sudo apt install rpi-update

Step 3. Download & Unzip My Github Repository

clone My GitHub repository

git clone https://github.com/thyagarajank/Object-Detection-with-Voice-Feedback-using-Raspberry-Pi-and-Tensorflow-Lite.git

Unzip the Folder "Object-Detection-with-Voice-Feedback-using-Raspberry-Pi-and-Tensorflow-Lite". Move the all files to "tflite1" folder.

uzip Object-Detection-with-Voice-Feedback-using-Raspberry-Pi-and-Tensorflow-Lite

Step 4. Create New Directory

Create New Dir at (/home/pi/)

mkdir tflite1

Move all Files to "tflite1"

mv Object-Detection-with-Voice-Feedback-using-Raspberry-Pi-and-Tensorflow-Lite tflite1

Work in this "/home/pi/tflite1" directory.

cd /home/pi/tflite1
pwd

Step 5. Create a Virtual Environment called "tflite1-env".

Intall Virtualenv

sudo pip3 install virtualenv

Create the "tflite-env" Virtual Environment

python3 -m venv tflite1-env

Activate Virtual Environment

source tflite1-env/bin/activate

Step 6. Install TensorFlow Lite & Other Dependencies.

OpenCV required packages and other libraries mention in .sh (shell) script. it will automatically download and installed.

bash install_tflite_requirements.sh

Step 7. Download Google's sample TFLite model

coco_ssd_mobilenet_v1_1.0_quant_2018_06_29 object detection model run on TensorFlow Lite

wget https://storage.googleapis.com/download.tensorflow.org/models/tflite/coco_ssd_mobilenet_v1_1.0_quant_2018_06_29.zip

Unzip the download folder.

unzip coco_ssd_mobilenet_v1_1.0_quant_2018_06_29.zip -d Sample_TFLite_model

NOTE: If use own custom detection model. Do not Download Google's SSDLite-MobileNet-v2 object detection model(step 7). Copy your own custom Model File to Sample_TFLite_model.

Lesson 2 - Run Object Detection in TensorFlow Lite

Checklist

• 1. Setup your webcam or Picamera plugged in
• 2. Enabled camera interface in Raspberry Pi (Click the raspberry icon in the top left corner of the screen, select--> Preferences --> Raspberry Pi Configuration, and go to the Interfaces tab and verify Camera is set to Enabled. After reboot the Raspberry Pi.)
• 3. Closing applications you aren't using and free up memory.
• 4. Before running the command, make sure the tflite1-env environment is active. (tflite1-env) appears in front of the terminial. This command for run object detection model only.

Step 8. Run the TensorFlow Lite Object Detection Model Only

Webcam Image object Detection Python Program.

python3 object_only_webcam.py --modeldir=Sample_TFLite_model

Step 9. Install Text to Speech Packages.

• 1. pyttx3
• 2. espeak Intall "pyttsx3" Text to Speech (TTS) library for Python 2 and 3

pip install pyttsx3

Install espeak library

sudo apt-get install espeak

Issue in install espeak Package

Try this 2 Command

sudo apt update && sudo apt install rpi-update

sudo apt-get install python-espeak

Step 10. Run the TensorFlow Lite Object Detection Model with Voice Feedback.

This Python Program for run object detection with voice feedback.

python3 object_voice_webcam.py --modeldir=Sample_TFLite_model

Lesson 3 - Train own Custom Object Detection Model

https://github.com/thyagarajank/tensorflow-own-custom-object-detection-model-in-google-colab Tensorflow (.pb file) Custom Model Training In colab (Code Availble in My Github). Tensorflow(.pb) file to Tensorflow Lite (.tflite) file Convertion Program (Upload Soon).