Cooper Cooperative Perception for Connected Autonomous Vehicles based on 3D Point Clouds

2022-11-30

Basic Information

Authors Qi Chen, Sihai Tang, Qing Yang and Song Fu

Organization University of North Texas

Source 2019 IEEE 39th International Conference on Distributed Computing Systems (ICDCS).

Source Code https://github.com/Aug583/F-COOPER

Content

Motivation

It is easy to determine that some detection failures are caused due to objects being blocked or existing in the blind zones of the sensors. Detection failures could also be caused by bad recognition because the received signal is too weak or because the signal is missing due to system malfunction.

Connected autonomous vehicles (CAV) can share their collected data with each other leading to more information. We propose that information sharing can improve driving performance and experiences.

Dataset

KITTI, a dataset contains labeled data that allows for autonomous vehicles to train detection models and evaluate detection precision. This paper use the 3D Velodyne point cloud data provided by the KITTI dataset.
Self-made dataset, two car with necessary sensors, such as LiDARs (Velodyne VLP-16), cameras (Allied Vision Mako G-319C), radars (Delphi ESR 2.5), IMU&GPSes (Xsens MTi-G-710 kit),

Baselines

Cooper and Single shot

Result

Figure 1. Number of cars detected and the detection scores in four KITTI scenarios.

Method

Since sharing all collected data is also impractical, we need to take into consideration the bandwidth and latency of vehicular networks.

By exchanging LiDAR data, local environment can be reconstructed intuitively by merging point clouds into its physical positions. In order to reconstruct local environment by mapping point clouds into physical positions, additional information is encapsulated into the exchange package. Said package should be constituted from LiDAR sensor installation information and its GPS reading, which determines the center point position of every frame of point clouds.