A Distributed Data Collection System for Traffic Images

Abstract

With the development of intelligent traffic system, high-definition cameras are spread along the urban roads. These devices transmit real-time captured images to data center for multi-purpose usability, but these bring higher requirements on network and storage capacity of the traffic images collection system. To address these problems, we proposed a compressed representation method for traffic images and collection system architecture. Firstly, the method proposed in this paper designed a distributed data collection system for traffic images based on edge computing mode. Secondly, we studied on the image feature representation methods for vehicle type/version retrieval, and formed a compressed representation method based on structural relationships selections. In this method, the retrieval precision reaches to 97.78% with the recall ratio of 90%, which proved the usability in this image collection system. Finally, we set up an analysis model based on Petri-net to observe the system requirements on storage, computing and transmission with different setting parameters. This model is powerful on finding bottlenecks of system in early stage and keeping balance in multi-aspects. The simulation experiments show that the data volume needs to be transported and preserved was compressed to 1/2250 comparing to the method of original images and the system transport delay was reduced more than 1/9 of original method. The experimental result showed that compared with the original collection method, the amount of data to be transmitted and stored was compressed by 1/2250, and the system transmission delay of the system was reduced to 1/9.15. This distributed data collection method and system proposed in this paper provided a novel referable revolution for traffic images processing system in intelligent traffics.