DeepAutoMapping: low-cost and real-time geospatial map generation method using deep learning and video streams

Abstract

Field data collection and geospatial map generation are critical aspects in different fields such as road asset management, urban planning, and geospatial applications. However, one of the primary impediments to data collection is the availability of spatial and attribute data. This issue is aggravated by the high cost of conventional data collection and data processing methods and by the lack of geospatial data collection policies. This study proposes an inexpensive approach that enables real-time field data observation and geospatial data generation from video streams connected to a laptop and positioning sensors using deep learning technology. This proposed method was evaluated via an application called “DeepAutoMapping”, which was built on top of Python, then underwent through two different evaluation scenarios. The results demonstrated that the proposed approach is quick, easy to use and that it provides a high detection accuracy and an acceptable positioning accuracy in the outdoor environment. The proposed solution may also be considered as a pipeline for efficient and economical method of geospatial data collection and auto-map generation in the future.