Situational Awareness with Ubiquitous Sensing: The Case of Robust Detection and Classification of Targets in Close Proximity

With the ubiquity of powerful mobile computational and communications devices (e.g., smart-phones, vehicles), technical capability that enables the utilization of ubiquitous sensing for detection and classification of ‘targets' or ‘events' of interest is very likely to create a paradigm shift in the situational awareness domain. While the use of machine learning methods have achieved reasonable success in certain detection and classification tasks, difficult situational awareness tasks, such as close proximity target detection and classification still remains as a challenging and unresolved problem. In this paper, a multimodal fusion and belief revision approach is presented with a particular emphasis on robust detection and classification of close proximity targets. By viewing classifiers as logical sensors, measurement modeling, sensor alignment and fusion steps are derived based on classifier characteristics (e.g., scope of expertise, performance indices). With a particular application of ubiquitous road hazard detection in place, the presented modeling approach is evaluated based on a hypothetical scenario that mimics ubiquitous detections corresponding to a real-life multiple hazards in close proximity. Furthermore, a real-life hazard detection scenario is also illustrated using the presented approach. In conclusion, a discussion is provided on the choice of fusion parameters and their impact towards robust detection and classification of hazards in close proximity.