Detecting Street Signs in Cities Based on Object Recognition with Machine Leaning and GIS Spatial Analysis

Abstract

Road traffic signs management is a process that searches, maintains, and builds traffic signs to ensure a normal functioning of traffic systems. Automatic road traffic signs detection is an important feature in smart cities. Existing road assets management systems usually rely on labor-intensive site inventory. Some other approaches use computer vision techniques to recognize traffic signs. Recent approaches combine GPS data and vehicle-based image recognition system to detect traffic signs along with geographic information. This research provides an innovative way to detect traffic signs based on geotagged photos from Google Street View. We used the Single Shot Multi-Box Detector based on a TensorFlow framework to train the recognition model. This process is implemented on a graphic card with CUDA acceleration to speed up the training process. Results showed that stop signs at road intersections can be accurately detected over 99%. This research helps to reduce workload for traditional traffic asset inventory. Our workflow can be used to detect other traffic signs and applied to other cities.