In the Real-time object detection on Raspberry Pi 4b using Tensorflow Lite article posted last month, I mentioned how I implemented real-time object detection on the Raspberry Pi. This week I managed to make it wearable. There are four parts in this wearable set: a box integrating battery and motherboard, a remote control, a pair of eyeglasses with a camera, and a pair of headphones.
I ran into a few problems during the production process. One of the biggest problems was that I tried to get the object detection program to run as soon as the Raspberry Pi boots, but it failed. Finally, I found this article and solved the problem: How to Run a Raspberry Pi Program on Startup - SparkFun Learn. It turned out to be because I had moved the location of the model file I was using in order to make it easier to find it, resulting in the system program not being able to find the location of the model at boot time. I changed the path of the model file to an absolute path in the source code and then the problem was solved. The underlined code in the image below is the modified one.
Here's a prototype of this device. The battery, the motherboard and the remote control are placed in a fanny pack.
In the future, I want to productize it further. Here are some design sketches.
Comentários