Nicolas Modrzyk - Real-Time IoT Imaging with Deep Neural Networks: Using Java on the Raspberry Pi 4

Arduino

In the world of dedicated hobbyists, Arduino was a breath of fresh air. It created an open source, free, unified ecosystem of libraries and connectors to plug things together. Users of the platform exchanged and competed for ideas in newly created labs within universities or renovated factory buildings.

I remember many years ago going to my friends open source, open hardware lab in Shanghai where they were moving robots around and re-creating arcade game rooms with old-school games.

While that was fun for real geeks, others may have felt they were left out a bit. The learning curve was just a bit high, and the power of Arduino made it so that running anything but C code was not practical.

Then, from England came the Raspberry Pi, a full-blown computer that could be held in your hands. At around $40 USD, it was an expensive purchase for students, but it provided some massively capable hardware. Hobbyists bought tons of those little devices. The Raspberry Pi 2 was quite a big jump in specifications compared to the Arduino. The biggest difference for the end user, aside from the connectors, was that you could finally run a full-blown operating system and get some standardization between the different components. Raspberry 3, especially model B, had 1Gb of memory, meaning you could run Java applications nicely with good performance. The underlying ARM-based processor was still quite limited in real-time computation, though.

Fast-forward to 2019: the Raspberry 4, a small computer with the same size and tag price, is a beast. In fact, Japan, not an EU member yet, bought a license in September 2019 to officially make the device usable within its borders. The Japan government went through the effort mostly because the Wi-Fi and Bluetooth modules embedded in the device allow for military-level spying and should be monitored carefully. Just kidding. But some extra checks are still officially required on those small devices.

So, whats new with version 4? The main difference is power, both CPU and memory-wise; it also has various video-rendering upgrades. But isnt that what every salesperson will tell you to get you to buy their latest and greatest version?

Well, yes and no. While in previous versions you could run Java processes and enjoy quite a bit of speed, the Raspberry Pi 4 is fast enough and powerful enough that you can run real-time object detection on real-time video streams. That is a game-changer and the very reason for this book: it is now possible to have some video fun by simply using the Java wrappers for OpenCV.