Tensorflow Lite Object Detetion on Raspberry PI

To run the tensorflow lite model on your raspberry pi, do follow the below steps:

Hardware Requirements

• Raspberry Pi
• PI Camera Module

Clone

Clone the repo onto your raspberry pi machine

git clone https://github.com/AnuragAnalog/TF-Object-Detection-on-PI
cd TF-Object-Detection-on-PI

Download the required packages

Run the download.sh file to download all the required packages which are necessary to run opencv

./download.sh

If you want to install tensorflow instead of tflite_runtime, you can comment the last four lines in the script.

Download the TensorFlow Lite Pre-trained model

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

After downloading the model, you need to unzip it, use the below command for that to unzip and store it in a directory called model

unzip coco_ssd_mobilenet_v1_1.0_quant_2018_06_29.zip -d model

Code

I have written two varaints of the program, one which takes input from your wecam feed and the other which takes input from a video file.

For video.py I have added a sample video file

Run the program

To run the webcam version

./webcam.py --model_dir=model

To run the video version

./video.py --model_dir=model

Demo

TODO List

• Object Detection on Webcam
• Object Detection on Video
• Object Detection on Image
• Add more links to pre-trained models
• Object Detection using Custom models

Problems

I have took atmost care in making this hassle free, but still if you face any issues, feel free to create issues.