A Deep Learning Framework for automated collection and analysis of traffic data based on identifying and classifying Delivery Vehicles in Logistics

Abstract

In urban logistics, analyzing urban traffic data plays an important role in achieving higher schedule reliability and delivery time efficiency. To increase the diversity of urban traffic data, we developed a solution for the automated collection and analysis of different types of traffic data. These are needed to optimize and control the flow of traffic. The use of traditional on-road sensors (e.g., inductive loops) for collecting data is necessary but not currently sufficient because it cannot draw any conclusions about the type of goods being transported. In this paper, we propose a framework in which different classes of delivery vehicles and types of goods being shipped are identified in road videos by deep-learning-based image recognition method. Video sequences are automatically evaluated according to the following criteria: (i) distinguish between individual and commercial vehicles, (ii) identify the category of commercial vehicles, for example, van, box trucks, small trucks, etc. (iii) identify the special features of the vehicle body (such as the name of the carrier) to classify commercial transportation of food, general goods or package services, etc. Using this method, logistics throughput of a designated city or region and the peak time of goods transportation can be obtained. This provides the carrier with better pre-advice and potential actions to improve transportation efficiency. For the evaluation of our framework, we collected real street videos at different time points in the main traffic arteries of Heilbronn, Germany. In particular, the difference between traffic flow of logistics services before and during the COVID-19 epidemic was compared. The results of implementation and testing demonstrated a high-precision, low-latency performance of the framework for obtaining urban logistics data.