Surveillance systems that use machine learning (ML) algorithms to identify people and objects are becoming increasingly widespread across the glober. However, new research by a group of Belgium academics highlights the significant shortcomings that still affect these systems, as the researchers were able to avoid detection by surveillance algorithms simply by attaching a 2D image to their clothing.
The researchers proceeded on the assumption that surveillance systems could potentially be tricked into misidentifying humans or objects if these were partially covered by images showing something else. After some experimentation, the researchers discovered that images of random objects that had been visually distorted were most successful at deceiving the algorithms. For the experiment to work, the image needed to be placed in the center of a person’s ‘detection box,’ so in front of the abdomen of someone standing straight and facing the camera, and it had to fully visible by the camera at all times.
While the technique is not perfect, it demonstrates how simple it can be for people to make themselves invisible to modern surveillance systems.
Read more: Academics hide humans from surveillance cameras with 2D prints
