Skip to main content
ICTD 2024 Main banner
Icon Legend
This session is not in your schedule.
This session is in your schedule. Click again to remove it.

Abstract for presentation (Poster or Podium) with a Paper in the Conference Proceedings

AI in Transportation

VEHICLE-TO-VEHICLE CONFLICTS DETECTION BY EMPLOYING KEYPOINT DETECTION ALGORITHM AND MULTI-CAMERAS

Monday, June 17, 2024
1:30 PM - 2:45 PM ET
Room: Galleria
Abstract: Over the last decade, researchers have demonstrated considerable interest in traffic monitoring, employing a variety of devices such as surveillance cameras, LiDARs, loop detectors, drones, and sensor fusions. The integration of these devices with computer vision applications has enhanced the analysis of traffic flow at intersections. Progress in computer vision has empowered traffic safety researchers to scrutinize microlevel details of traffic scenes, enabling more thorough assessments. These techniques have found widespread application in vehicle detection and tracking, leading to the establishment of robust safety assessment methodologies.
Despite their efficacy, the widespread implementation of these techniques faces limitations nationwide. To overcome these challenges and optimize safety assessment, this study introduces an innovative approach that combines surveillance cameras with a carefully crafted framework. This approach integrates the fields of views of multiple surveillance cameras and employs two distinct CNN-based detection algorithms to identify and track vehicles, yielding precise and dependable outputs. These outputs undergo further analysis, smoothing, and integration to reconstruct the entire scene, facilitating the extraction of crucial traffic flow features for vehicles at signalized intersections.

This framework offers a comprehensive solution for traffic safety assessment by calculating traffic conflict indicators, and identifying instances of traffic conflicts. This integrated approach provides an efficient and accurate safety assessment tool. By leveraging the strengths of the integration between surveillance cameras and CNN-based detection algorithms, this research aims to contribute significantly to the advancement of traffic safety practices.

Learning Objectives:

Attendees can expect to learn the following from this session:
  • Upon completion, participants will be able to integrate the fields of views from multiple cameras to provide more accurate and wider fields at signalized intersections to monitor vehicular movement.
  • Upon completion, participants will be able to utilize two convolutional neural networks-based algorithms, YOLOv7 and OpenPifPaf, to detect vehicles by bounding boxes and key points.
  • Upon completion, participants will be able to cluster the extracted key points from multiple cameras, determine center points, and set average dimensions for each detected vehicle after transformation to plan-view.
  • Upon completion, participants will be able to extract vehicles trajectories, smooth them, and calculate Time-to-Collision as a traffic conflict indicator. Then, identify traffic conflicts instances.