Attention Based Framework for Traffic and Collision

This paper presents a solution aimed at mitigating the risk of traffic collisions and automating immediate emergency response when such incidents occur. Our approach involves cascading an object detection model, supplemented with a tracking algorithm and a prediction model based on attention. To sel...

Full description

Saved in:
Bibliographic Details
Published inIEEE International Conference on Communications (2003) pp. 1 - 4
Main Authors Popescu, Constantin-Bogdan, Florea, Corneliu
Format Conference Proceeding
LanguageEnglish
Published IEEE 03.10.2024
Subjects
collision
Detectors
Emergency services
object
Object detection
Particle measurements
Prediction algorithms
Predictive models
Proposals
traffic
Trajectory
Transformers
visual transformer
Online AccessGet full text

Cover

More Information
Summary:This paper presents a solution aimed at mitigating the risk of traffic collisions and automating immediate emergency response when such incidents occur. Our approach involves cascading an object detection model, supplemented with a tracking algorithm and a prediction model based on attention. To select the appropriate detector a comparison between recent proposals is implemented. A secondary network is then employed to forecast the future positions of traffic participants. With these predictions, we calculate the potential for contact among the objects, providing a predictive measure of collision risk. The proposed solution is evaluated on a relevant database and the best performers are identified.
ISSN:1938-1883
DOI:10.1109/COMM62355.2024.10741460