Novi Sad, Serbia – 2018 marks the third anniversary for ZINC events, which include the Zooming Innovation in Consumer Electronics International Conference 2018 (ZINC 2018), which is sponsored by IEEE Serbia and Montenegro section and CE chapter and RT-RK Institute and RT-RK department at University of Novi Sad.
ZINC Events provide symposiums, forums, seminars, talks, workshops, tutorials, product and prototype demonstration, for researchers and engineers from academia and industry to exchange experience and results on their recent work.
The Visage Technologies team, consisting of Dijana Vitas, Martina Tomic and Matko Burul, was glad to participate at the conference and present an article on the subject of “Image Augmentation Techniques for Cascade Model Training”.
The paper investigates the influence of different augmentation methods on traffic light classifiers training. Image augmentation is the process of taking images and manipulating them in order to create many altered versions of the same image. This provides more images to train on, but can also help expose classifier to a wider variety of lighting and coloring situations to make a classifier more robust.
Some augmentation methods affect sample’s color and illuminance characteristics, such as brightness, contrast modifications and color jittering based on principal component analysis. Additionally, the article investigates noise, blur as well as horizontal flip methods.
Each method is applied to a set of training data and the impact it had on the training process is analyzed. Results are presented in the form of true positive and false positive numbers obtained when using different augmentation methods.
For more info about the conference, please visit the ZINC website.