Crowd Map: Accurate Reconstruction of Indoor Floor Plans from Crowdsourced Sensor-Rich Videos

Abstract

Lack of an accurate and low-cost method to reconstruct indoor maps is the main reason behind the current sporadic availability of digital building floor plans. The conventional approach using professional equipment is very costly and only available in the most popular areas. In this paper, we propose and demonstrate CrowdMap, a crowd sourcing system utilizing sensor-rich video data from mobile users for indoor floor plan reconstruction with low-cost. The key idea of CrowdMap is to first jointly leverage crowd sourced sensory and video data to track user movements, then use the inferred user motion traces and context of the image to produce an accurate floor plan. In particular, we exploit the sequential relationship between each consecutive frame abstracted from the video to improve system performance. Our experiments in three college buildings show that CrowdMap achieves a precision of hallway shape around 88%, a recall around 93% and a F-measure around 90%. In addition, we achieve on average 9.8% room area error and on average 6.5% room aspect ratio error. The evaluation result demonstrates a significant improvement of accuracy compared with other crowd sourcing floor plan reconstruction systems.