Machine learning to measure the nettovolume of logs in piles

Purpose and goal

The purpose of the project was to evaluate the possibilities to use machine learning algorithms to increase the automation of measurement of piled logs on lorries. The idea was to use available historic data, both high resolution images and 3D reconstructions of trucks with piled logs in combination with measurement results from manual measurement in order to select and train machine learning algoritms to estimate the netto volume with sufficient accuracy.

Expected results and effects

The conclusion of the project is that it should be possible to use a combination of different machine learning algorithms on both the 3D reconstructions and images in order to measure the netto volume with an accuracy comparable with the current manual measurement methods. The approaches and algorithms evaluated in the project will most likely be implemented in a future product.

Planned approach and implementation

First a general understanding of the domain, i.e. the different parts of a correct netto volume estimation, was understood and the existing data was understood and structured in order to be used for training and verification. Different machine learning algorithms was the trained and evaluated for the different parts, using one set of the data for training and one for verification. Some of the algorithms were also verified in a scale model of the product.