A spatiotemporal data fusion model for occupancy state estimation: an evidential approach

In the scope of the EU funded TRACKSS project on cooperative advanced sensors for road traffic applications, we investigate the potential of pre-existing road traffic sensors for pedestrian crossing detection. Two road traffic video-sensors provide spatial occupancy rates on areas along a crosswalk. We propose to correct pedestrian under-detection with a double fusion process composed of inter-sensor and intra-sensor fusion. Both are framed in the transferable belief model. The intra-sensor fusion uses the spatiotemporal characteristics of the occupancy rates with two steps: past beliefs update according to an estimation of ongoing spatiotemporal evolution, and temporal data fusion. The results obtained on real data collected on an urban intersection show that both fusion processes - and especially the intra-sensor fusion - contribute to significant improvements in occupancy state estimation, and enable to reach high crossing detection rates in operational conditions.