Research on Vehicle Re-Identification Based on Visual Features in Intelligent Express Way System

Theme

Image and Video

Topic

Research on Vehicle Re-Identification Based on Visual Features in Intelligent Express Way System

Background

Express way toll system in China is evolving fast and nowadays it has become automated with fees being collected when a vehicle passing by a toll gantry, which captures the passing signal of the vehicle, in two ways, electrical and visual images respectively. It has been found that in practice the electrical signal itself can be invalid at which time the visual images become an important information source for trans-fee justification. 

In such case, Vehicle re-identification (ReID) becomes a vital task to recognize different vehicles, though maybe of the same type, to reconstruct vehicle trajectory. Under normal circumstances, we can reconstruct vehicle trajectory by combining vehicle images with the same license plate number. However, when exceptions such as fake plate, plate occlusion, or temporary license plate occur, it is difficult to reconstruct the correct trajectory only with license plate numbers. Therefore, research on vehicle ReID based on visual features along with plate number recognition can play a key role in identifying different vehicles in practice.

The goal of this project is to establish a vehicle ReID method, restricted to the expressway scenario, with real-world application accuracy. Vehicle ReID problem faces challenges including high intra-class variability caused by the dependency of shape and appearance on viewpoint and weather, and small inter-class variability caused by the vehicles of the same style produced by the same manufacturers, etc., which usually leads to low accuracy results in practice. A vehicle ReID research in the expressway scenario is demanded, with high accuracy, satisfying recall rate and high efficiency to handle large amount of visual data. 

The researchers will be able to access our data set with more than 10million vehicle images and the corresponding time and location information. The applicants are encouraged to do any ReID related researches with the provided dataset, as long as the developed technology can lead to better utilization of the information of these images or promoting the computer vision technology in real practice scenario.

Targets

  • A vehicle ReID model to solve the problems of vehicle ReID in real expressway scenario efficiently, including model source code, performance reports, etc.
  • ≥1 paper published in the CCF recommended catalogue or Chinese Academy of Sciences Journal Division Table 3 or above; or JCR T3 and above.

Related Research Topics

  • Salient feature selection and feature embedding of vehicle in expressway
  • context embedding for vehicle Re-ID
  • Fusion of vehicle feature and context for vehicle ReID
  • Large-scale vehicle ReID in real expressway scenario.
  • performance optimization of vehicle ReID in large scale image data set.

 

Suggested Collaboration Method

AIR (Alibaba Innovative Research), one-year collaboration project. 

 

Scan QR code
关注Ali TechnologyWechat Account