Mind you, the year was 2017. I trained a deep convolutional neural network for brand logo detection using the darknet deep learning framework (now taken over by hank.ai), in C.

For deployment of the model, I wrote an Android app that captures images from a USB webcam and feeds the images to the neural network.

Making Of

1. Get Glasses

2. Attach Webcam

3. Train, test, and select best model

Read the full story at NIM.org, my former employer!

Mobile logo detection in the wild

I created a video from everything the glasses "saw" during a stroll through the city of Nuremberg, Germany, with bounding boxes around all logos from our dataset.

The bounding boxes are green when the glasses recognized a logo correctly, red when a logo was detected but got the brand wrong, and white when the glasses missed a logo.

The ground truths were carefully annotated by our working student.

Here is the start frame of the video, with details of the task.

The results were pretty good:

A small casestudy revealed the following:

Here are a few sample detections: