Crowdsourcing soft data for improved urban situation assessment

Conventional “hard” sensing in urban spaces is challenged by the complexity of the environment, creating gaps in situation assessment and possible confusion due to data association errors. Crowdsourced “soft” reports from human observers may remedy this problem but require techniques for fusing hard and soft data. This paper describes an experimental crowdsourcing system to evaluate the potential improvement in situation assessment resulting from the fusion of hard and soft data. The paper then applies a new combination of Bayesian inference algorithms, Particle Filtering and Softmax learning, to a canonical test problem: tracking a single moving object moving along a road network. The fusion of soft reports with intermittent hard data is shown to yield a marked improvement in situation assessment performance. Key to achieving such gains in practice will be appropriate incentives to reward trustworthy reporters along with methods to reduce sensitivity to any remaining untrustworthy reports.