Klasifikacija delovnih operacij na osnovi posnetkov gibanja dlani

Zupanc, Anej (author), Vrabič, Rok (mentor)

V diplomski nalogi predstavimo grajenje modela za strojno učenje, ki bo služil za klasifikacijo delovnih operacij na podlagi posnetkov gibanja dlani. Pri tem uporabljamo knjižnico MediaPipe za zbiranje podatkov in knjižnico TensorFlow, Keras za strojno učenje. Slednje je izvedeno z uporabo ponavljajočih se nevronskih mrež, specifično dolgoročnega kratkoročnega spomina. Celoten model je zgrajen v programskem okolju Python. Na koncu predstavimo statistične rezultate zgrajenega modela, ki kaže 91,25% uspešnost.

• Language: Slovenian
• Keywords: klasifikacija delovnih operacij, posnetki gibanja dlani, MediaPipe, strojno učenje, dolgoročni kratkoročni spomin, Keras
• Work type: Final paper
• Typology: 2.11 - Undergraduate Thesis
• Organization: FS - Faculty of Mechanical Engineering
• Place of publishing: Ljubljana
• Publisher: [A. Zupanc]
• Year: 2021
• Number of pages: XIV, 30 str.
• PID: 20.500.12556/RUL-130218
• UDC: 004.85:005.935(043.2)
• COBISS.SI-ID: 79980803
• Publication date in RUL: 11.09.2021

Secondary language

• Language: English
• Title: Classification of work operations based on an analysis of hand movements

Abstract

In this bachelor's degree we present building of a machine learning model, capable of classification of work operations based on an analysis of hand movements. In doing so, we use MediaPipe library for keypoint and data collection and TensorFlow, Keras library for machine learning. The latter is done using recurrent neural networks, specifically long short term memory. The whole model in build in a Python software environment. At the end we present statistical data, showing 91,25% classification performance.

• Keywords: classification of work operations, videos of hand movements, MediaPipe, machine learning, long short term memory, Keras