A novel vehicle lateral positioning methodology based on the integrated deep neural network

Abstract

Many intelligent transportation system (ITS) applications increasingly depend on accurate and reliable positioning performance. How to achieve such performance using low-cost sensors is one of the main challenges for land vehicles. Traditional positioning sensor such as the Global Positioning System (GPS) may fail to obtain satisfactory performance in challenging situations. The main reason is that satellite signals along the vehicle's lateral direction are obstructed by obstacles and therefore the lateral position error will be increased. This paper proposes a novel methodology to achieve lane-level lateral positioning based on the integrated deep neural network (IDNet) which integrates the deep convolutional neural network (DCNN) with the multi-layer neural network. Specifically, IDNet consists of two carefully designed networks that are connected in series. The first one aims to extract the road area efficiently. The other is integrated with the former and makes full use of the road area to predict lateral position accurately. The design concept of IDNet is inspired by the mechanism of human eyes’ lateral positioning, which utilizes the location information hidden in road areas to predict the lateral position. Benefiting from this novel mechanism, the proposed methodology applies to many urban road scenes including well-marked, poorly marked and unmarked roads. To evaluate this approach, road experiments with typical scenarios were performed in urban streets. The results validate its effectiveness and reliability. From the methodological perspective, the proposed method simulates the process of human eyes’ lateral positioning, which indicates that artificial intelligence provides an effective means to simulate human perception and behavior for domain experts. From the application point of view, the proposed methodology expands the application fields of artificial intelligence, such as the field of positioning involved in this paper.