An Interactive Multi Model prediction-based Radar-Camera Fusion Algorithm for Autonomous Vehicles

Jinpeng Luo,Hui Xie,Kang Song
DOI: https://doi.org/10.1109/ccdc55256.2022.10033465
2022-01-01
Abstract:Continuous and reliable object detection is vital for environment perception, planning and control of autonomous vehicles. As most popular sensors, the millimeter-wave radar and camera both suffer from false detection in the presence of disturbances, such as electromagnetic interference for radar and fuzzy image for camera. Therefore, in this paper, an interactive multi model prediction-based radar-camera fusion(IMM-RCF) algorithm for autonomous vehicle is proposed to tackle the aforementioned issues. While detecting the position of the same target using camera and radar, confidence level of measurements from camera and radar are evaluated by the target’s predicted location from a compound model with a weighted sum of multiple kinematic sub models. A relatively more reliable signal, either from the camera or from the radar, can be easily selected by comparing the deviation between compound model estimates and the camera or radar measurements. This allows a reliable and fast distinguishing of detection result. To improve the accuracy of the prediction result, weighting factors for the output of each sub-model are updated by observing the estimation error of each sub-model. Additionally, for a smoother output, a radar and camera signals fusion algorithm is proposed following the philosophy of Kalman filter. Experimental results on an autonomous bus show that the IMM-RCF algorithm reduced the false detection rate to 1.86% also decreases measurement error to 0.60 m for objects 39.21 m ahead of the bus, demonstrating 53.4% and 54.4% improvements relative to radar and camera measurement results in distance accuracy.
What problem does this paper attempt to address?