Skip to content

TeRyZh/511-CCTV-Camera-Vehicle-Detection-and-Tracking-Using-STMap-and-DAIU-Net

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

67 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

511 CCTV Camera Vehicle Detection Using DAIU Net and STMAP For Advanced Traveler Information System (ATIS)

Zhang T, Jin PJ, Ge Y, Moghe R, Jiang X. Vehicle Detection and Tracking for 511 Traffic Cameras With U-Shaped Dual Attention Inception Neural Networks and Spatial-Temporal Map. Transportation Research Record. 2022;2676(5):613-629. doi:10.1177/03611981211068365

Scanline Method for NJ511 Traffic Camera Network

Highlights

  • Proposed Dual Attention Inception Neural Network for traffic detection that has faster converge speed and better performance to high-impact image segmentation Models
  • An Efficient STMap-based Vehicle Detection Model Under Various Illuminations, Infrastructure Noises, Weather Conditions better than the SOTA traffic video understanding methods

Abstract

This project developed a Spatial-temporal Map (STMap) based vehicle detection method for 511 camera networks as an add-on toolbox for the traveler information system platform. The U-Shaped Dual Attention Inception (DAIU) deep learning model was designed, given the similarities between the STMap vehicle detection task and the medical image segmentation task. The inception backbone takes full advantage of diverse sizes of filers and the flexible residual learning design. The channel attention module augmented the feature extraction for the bottom layer of the UNet. The modified gated attention scheme replaced the skip-connection of the original UNet to help the model to reduce irrelevant features learned from earlier encoder layers. The designed model was tested on NJ511 traffic video for different scenarios covering rainy, snowy, low-illumination, and signalized intersections from a key, strategic arterial in New Jersey. Based on segmentation model evaluation metrics, the DAIU net has shown better performance than other mainstream neural networks. The DAIU based STMap vehicle detection is also compared against the state-of-the-art solution for infrastructure-based traffic video understanding and demonstrates superior capability. The code for the proposed DAIU model and reference models are made public, and the labeled STMap data to facilitate future research.

Vehicle Detection Results

Compared to AI City Challenge Winning Solution

License

The source code is available only for academic/research purposes (non-commercial).

Contributing

If you found any issues in our model or new dataset please contact: [email protected]

About

VEHICLE DETECTION AND TRACKING FOR 511 TRAFFIC CAMERAS WITH DAIU AND STMAP

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published