Affiliation:
1. Tongji University, China
Abstract
<div>The closet in-path vehicle (CIPV) is recognized relying on the detection results
for road lane lines in most current ACC system, which may not work well in the
poor conditions, for example, unclear road lane lines, low light level, bad
weather, and so on. To solve this problem, the article proposes a sensor
fusion-based CIPV recognition algorithm independent of road lane lines. First, a
robust Kalman filter based on the global coordinate system is designed to fuse
the millimeter-wave radar and camera targets. The fusion algorithm can
dynamically adjust the covariance matrix of sensor observations to avoid the
influence of anomalous observations on the fusion results. Stable detection of
targets by the fusion algorithm is the basis of the CIPV recognition algorithm.
Then, the CIPV recognition algorithm generates virtual lane lines using the
motion parameters of self-vehicle or the driving trajectory of vehicle target
and develops a mode switch strategy for virtual lane lines generation based on
the driving state of target. This strategy can flexibly switch to the applicable
virtual lane lines generation method in different scenarios. Finally, field
tests are conducted in typical scenarios to verify the performance of the CIPV
recognition algorithm. The results show that the algorithm is able to recognize
CIPV stably and accurately without relying on road lane lines.</div>
Subject
Artificial Intelligence,Computer Science Applications,Automotive Engineering,Control and Systems Engineering,General Medicine