As the use of autonomous systems in solving complex problems increases, so too does the demand for decision-making models capable of adapting in response to dynamic environments in real time. Suitable inference techniques in computer vision to achieve most decision enabling tasks exist, and the visual domain provides a rich source of data that can be used to inform agents. Time, however, is a critical constraint: existing inference techniques typically require a large number of computational resources and cannot produce decisions rapidly enough to enable agents to keep pace with human perception. The primary focus of my research is to address the requirement for real-time analysis of visual input enabling decision making in autonomous systems in addition to ways lower inference time can be exploited in the creation of new data modelling techniques.