Motorcycle detecting mirror

How this project came about

There's a story behind why I wanted to make a mirror that helps car drivers sense when there's a motorcycle just behind them - a painful story. "Multiple broken ribs (front and back) and a shattered shoulder blade" sort of painful. Yes, you guessed it. A little over a year ago I was riding my motorcycle along, travelling in traffic just beside a car and the driver decided that the traffic he was in was too slow and suddenly U-turned.

Maybe it was because I was on a quiet electric motorcycle that he didn't know I was there. We were both moving slowly and I was just to the side of his car but I would have been visible in his mirror if he'd looked. Whatever the reason, I wish he had known I was there. It would have saved me 3 days in hospital and many months recovery. It would also mean I'd still have my Zero FXS electric bike.

A friend was showing me his new (also electric) car and the mirrors have a lovely red LED inset to warn him of obstacles detected when he's reversing. Wouldn't it be great if a driver could also be warned of thing to the side when driving in traffic?

Some cars already have a "blind spot warning" but this is only really effective for larger objects and more suitable when travelling at speed on a motorway. They don't really work for motorcycles in slower traffic.

Choice of components

I originally intended to use an Ultra96-V2 alongside the MIPI adapter mezzanine and a Pcam 5C camera. However despite this being an official 96boards product, these only seemed to have ever been available via eBay - and sadly were no longer available at all.

Luckily, just as I discovered this the Avnet/96Boards ON Semiconductor Dual Camera Mezzanine was announced and I snapped one up as soon as they were available. This looks like a perfect camera module for doing object detection!

Problems with the MIPI adapter mezzanine

The new MIPI adapter mezzanine might be perfect for vision applications on the Ultra96 but software support is not really there yet. At the time of writing, you need to get some beta software from Avnet and take a look at Adam Taylor's recent blog about the board before you can really get started. There's a lot of IP needed to get to a usable video stream in Vivado and I'm yet to try to direct this towards Vitis AI.

Problems using Vitis AI

I decided the easiest way to get started with traffic detection would be to start from the examples outlined by Mario Bergeron here on Hackster.io. https://www.hackster.io/AlbertaBeef/vitis-ai-1-2-flow-for-avnet-vitis-platforms-7cb3aa

Unfortunately, despite being fairly familiar with Xilinx tools I really struggled to get this compiling on an Ubuntu 18.04 VM that I set up specifically to run Vitis AI.

Stalled project?

Unfortunately this project is proving very difficult to get started,