20,290
edits
Assembly Line Activity Recognition: Difference between revisions
m
→Project context
| Line 19: | Line 19: | ||
As part of RidgeRun's intention to expand its solutions portfolio, the company is venturing into the field of machine learning with a special focus on deep learning applications for the industry such as activity recognition. | As part of RidgeRun's intention to expand its solutions portfolio, the company is venturing into the field of machine learning with a special focus on deep learning applications for the industry such as activity recognition. | ||
Industrial processes are great candidates for activity recognition based on machine learning and computer vision because it occurs | Industrial processes are great candidates for activity recognition based on machine learning and computer vision because it occurs in controlled environments, and the number of variables can be controlled and managed according to the task. So, RidgeRun partnered with a manufacturing company in the field of multimedia hardware to explore ways of automating the assembly validation process i.e. being able to automatically classify assembly actions so a further validation process can take place. | ||
In order to provide automated assembly validation solutions, RidgeRun is testing and researching multiple smart activity recognition techniques. This project explores the viability of the deep learning approach and also provides the tools and workflow required for future similar applications. | In order to provide automated assembly validation solutions, RidgeRun is testing and researching multiple smart activity recognition techniques. This project explores the viability of the deep learning approach and also provides the tools and workflow required for future similar applications. | ||
The manufacturing third party provides the environment for data capture and the problem to be solved while RidgeRun provides all the development resources to tackle the problem, ranging from data capture | The manufacturing third party provides the environment for data capture and the problem to be solved while RidgeRun provides all the development resources to tackle the problem, ranging from data capture to network training. | ||
This project's tools, methodology, and knowledge can be applied to many other industrial environments and problems in order to automate, validate, monitor, and improve manufacturing processes. The improvements in products and their manufacturing complexity call for new and more advanced validation techniques, these new requirements are the driving force behind this project's development. | |||
=== Problem description === | === Problem description === | ||
| Line 32: | Line 31: | ||
Currently, the manufacturing company doesn't have an automatic way to recognize the actions performed on the assembly line, which can produce defective assembled parts that affect the quality of the final product. | Currently, the manufacturing company doesn't have an automatic way to recognize the actions performed on the assembly line, which can produce defective assembled parts that affect the quality of the final product. | ||
The problem mentioned above can be addressed manually by monitoring the production line; however, this will result in | The problem mentioned above can be addressed manually by monitoring the production line; however, this will result in inefficient use of resources and relies on human detection and decision making, which could be biased and inaccurate. The implementation of an automatic recognition system allows efficient validation of the assembly process without human intervention. | ||
This project focuses on how to detect and recognize activities on an assembly line from video samples using machine learning and computer vision techniques. | This project focuses on how to detect and recognize activities on an assembly line from video samples using machine learning and computer vision techniques. | ||
| Line 38: | Line 37: | ||
=== Parts and workstation === | === Parts and workstation === | ||
The activities to be recognized are from an assembly line that utilizes six parts that are placed in a specific sequence by a worker in an assembly workstation. These parts are listed | The activities to be recognized are from an assembly line that utilizes six parts that are placed in a specific sequence by a worker in an assembly workstation. These parts are listed below: | ||
# Riveted star | # Riveted star | ||
| Line 48: | Line 47: | ||
The assembly occurs at a static workstation where the parts are laid out in a particular order, making this problem a classification task instead of a detection task. | The assembly occurs at a static workstation where the parts are laid out in a particular order, making this problem a classification task instead of a detection task. | ||
The following figure shows the workstation where the parts are assembled | The following figure shows the workstation where the parts are assembled and details how the parts are laid out. It also shows a number for each section of interest according to the following numbering: | ||
# Riveted star | # Riveted star | ||