You Don’t Want to Step to This

If you are looking to move anytime soon, I have some free advice for you — stay far, far away from Roni Bandini’s neighborhood. Over the past several months, Bandini’s free time has been consumed by building a number of electronic devices that are designed to deal with annoying (or dangerous) neighbors. There was the Reggaeton Be Gone, for example, that jams nearby Bluetooth speakers when someone decides to blast reggaeton music in the early hours of the morning. More recently, Bandini built the Step Guard to keep drunks and thieves from hanging out on his doorstep doing drunken and thieving things.

The most recent iteration of Step Guard, Step Guard 2.0, used some relatively low-tech sensing methods, like an ultrasonic distance sensor, to detect the presence of people before shooing them away with a loud alarm if they hung around too long. This approach generally worked pretty well, but over time, Bandini found that there were some clear drawbacks. In some cases, for example, a promotional flier taped to his door would trigger the distance sensor and cause the alarm to sound.

The simple approach just was not going to cut it any longer, so Bandini recently upgraded the hardware and named the new device … drum roll, please … the Step Guard 3.0. Clever, huh? The latest version leverages computer vision and machine learning to make sure that the alarm is only triggered in cases where there is a legitimate concern.

To make this work, Bandini used Seeed Studio’s Grove Vision AI Module V2. This device comes loaded with an Arm Cortex-M55 CPU as well as an Ethos-U55 machine learning processor to speed up AI inferences. It also supports popular machine learning development frameworks like TensorFlow and PyTorch to get projects off the ground quickly. This was paired with a Raspberry Pi camera, then a pre-trained person detection model was loaded onto the board.

This new module was integrated into the existing Step Guard system as a replacement for the ultrasonic distance sensor. Since the new method can detect people with precision, and not through a coarse measurement like a distance sensor will provide, it allows Step Guard to be triggered only when people are legitimately loitering on Bandini’s step. Without this, Bandini’s neighbors might have needed to build Step Guard Be Gone systems to shut down a Step Guard that went haywire from a flier that was taped to the door. That might be an interesting electronics arms race to report on, but is probably not good for the neighborhood.

A few other minor enhancements were added to Step Guard 3.0, so be sure to check the project write-up for the full details. Is this the end of the Step Guard saga? It is hard to be certain, but Bandini hints that more updates may be in the works.