Novelty detection is a key component of biological vision systems, where its role is to extract critical elements for the agents survival from the massive amount of information present in his visual environment. Current vision based embedded systems , such as surveillance cameras, are facing similar challenges as they have to handle a significant amount of sensory data, with limited computing power and memory bandwidth available. In order to perform artificial novelty detection in these systems, it is necessary to have a model able to learn the local visual environment without having any prior knowledge. This study explores bio-inspired unsupervised neural networks models, more precisely self-organizing maps, which are good candidates for this task. We present an original approach consisting of performing novelty detection based on vector quantization and topological learning.