Machine Learning, and more specifically Deep Learning, are being widely used in the industry to assist human operators in their labors. One of the main applications is the automated action recognition which helps companies to get production statistics, improve assembly quality, enforce safety, optimize processes, among other applications. For this reason, RidgeRun has been actively working on several action recognition projects targeted to provide the fundamental basis for similar applications in our customers’ scenarios.
In this research report, you will find exhaustive details on how to tackle a real life application. The use case presented is the automated detection of an assembly sequence of a hardware part in a real production line. The report will guide you through the process of acquiring the dataset, performing the corresponding exploratory data analysis (EDA), and finally training and tuning the model to achieve an acceptable recognition score.
The SlowFast architecture was chosen for the project and details about its performance when applied to new video samples are shared. With a final accuracy and F1 scores above 0.90, this trained model represents an excellent choice for similar applications.
Check out more about this exciting project at our developers wiki page!
Do you have a similar project in mind? Feel free to contact us at support@ridgerun.com