“Closing the Loop”

Source: Deep Learning on Medium

This week we got the arduino leds working and were able to send signals for specific leds to light based on objects recognized with YOLO. One led will light green when a “bottle” is detected and another will light red when a “cup” is detected. If both are detected both leds will light blue. If none are detected neither led will light.

In the future we will light each led based on whether we classify it as waste or as recyclable. We also retrained YOLO with the data set TrashNet which can be found on GitHub. Given that the data set was not tailored to our application it decreased the performance of the system. In order to increase performance through retraining we will likely need to create and label a data set ourselves.

After taking these photos the system now lights green when a bottle or cup is detected and red when an apple or banana is detected. These colors correspond to recycling and waste. We also smoothed the lights over 20 frames taking a weighted average over the last 20 frames and comparing it to the threshold we set.