A STUDY ON DETECTING VEHICLES IN DEVELOPING COUNTRIES USING R-CNN

Abstract

 In Southeast Asia, there are unique transport modes: LAMAT which stands for locally adopted, modified and advanced transport. Although these are rooted as the legs of local residents, they often have no prescribed route and stop. Due to insufficient registration and management by local governments, LAMAT's actual operation and distribution are not clear in some countries.
 We developed a tool to detect LAMAT's car body from the satellite image data such as Google Earth and to estimate their distribution in the city using the method of R-CNN (Region with CNN). We verified the accuracy and the robustness of the developed tool by applying the tool to tricycle and jeepney along the actual street in Makati, the Philippines. We could grasp some factors that contribute to accurate detection and classification of LAMAT.