volumetric

Dataset and code for volumetric estimation of non-structured heterogeneous waste. Supplement for Automation 2021 paper.

How to use?

1. Unzip all files in data folder.
2. Run the script volumeEstimation.py.

Measurement system description

Running several sensor in ROS to perform and record trash level detection in a bin.

Sensors:

• Radar TI IWR6843
• Radar A111
• Realsense D435i
• Ultrasound sensor (legacy)
• ToF sensor (legacy)

Connected to Raspberry Pi 4:

• Radar TI IWR6843 -> via USB (RS 232)
• legacy sensors -> via ADC HAT board
• Radar A111 -> via SPI HAT board

Connected to PC (NUC):

• Realsense D435i (USB 3.0)

Block diagram

Sensor module photo

Test station photo

Data example

Neural network training

Two approaches for neural network training were used:

1. Use object volume information as ground truth.
2. Use depth camera (RealSense) data as ground truth.

Neural network architecture

The following architectures of neural networks were used for volume estimation from A111 and IWR6843 radars.