How To Build Object Detection Application Using Tensorflow Lite and Raspberry Pi 4
Due to huge processing requirements, neural networks have traditionally been limited to cloud servers.
But yet, making them widely available on small IoT devices like Raspberry Pi and smartphones would literally make them smart.
In manufacturing, on-board machine learning capabilities for plant devices have more compelling implications, which include low latency, enhanced security or long-term cost-effectiveness.
Take, for example, a manufacturing plant requiring image classification services at each of its 100 work stations. Hosting the image classification models on the cloud may not be the best solution in such a case, as this would be expensive.
A cost-effective solution would be to train the models on the cloud and ship the trained models to be executed entirely on the edge devices. This would also allow devices to be customised for detecting the set of objects unique to their workstation.
But, to do that, you’d require a toolkit that enables you to convert and optimise ML models to run on devices with constrained processing power. And that’s where Tensorflow Lite comes in.
In the video below, I demonstrate how to use Tensorflow Lite to build an object detection application that runs on Raspberry Pi 4.
THE HALF LIFE OF A LEARNED SKILL USED TO BE 30 YEARS. TODAY IT'S 5.
And it will continue to decrease exponentially. The only way to avoid becoming irrelevant is to get in the habit of reinventing yourself every single day.