Object Detection in Rural Roads through Single Shot Multibox Detector Mobilenet Network
Abstract— The object detection in the rural roads is an important perception module in any advanced driver assistance or autonomous driving system. Traditional techniques work reasonably well for this problem (urban areas) when the roads are well maintained, and the boundaries are clearly marked. In this paper, we propose a used deep-learning framework such as Single Shot MultiBox Detector (SSD) Mobilenet (retrain and pre-train) and GoogLeNet (pre-train) for object detection in a rural road. The deep learning neural networks were evaluated with 12 videos with a duration of 60 minutes to the training process. The result shows the best performance with the SDD Mobilenet and the confidence score 0,5 of un-maintained rural road commonly found in developing countries as well as rural areas in the developed world. With the previous result and making new testing used 4 videos, our SDD Mobilenet achieve improved car detection (PRE 73%, REC 76 %, F1 64%) and pedestrians detection (PRE 85%, REC 80%, F1 64%).
Index Terms— Object detection, computer vision, deep learning framework, processing image
Universidad Técnica Particular de Loja, Departamento de Ciencias de la Computación y Electrónica, SPAIN
Luis Barba-Guaman, José Eugenio Naranjo
Universidad Politécnica de Madrid, Instituto de Investigación del Automóvil, SPAIN
Cite: Luis Barba-Guaman, José Eugenio Naranjo, "Object Detection in Rural Roads through Single Shot Multibox Detector Mobilenet Network," Proceedings of 2019 the 9th International Workshop on Computer Science and Engineering, pp. 227-232, Hong Kong, 15-17 June, 2019.