Trailer angle estimation using radar point clouds

• Published in: Signal processing Vol. 188; p. 108221
• Main Authors: Olutomilayo, Kunle T., Bahramgiri, Mojtaba, Nooshabadi, Saeid, Oh, JinHyoung, Lakehal-Ayat, Mohsen, Rogan, Douglas, Fuhrmann, Daniel R.
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.11.2021
• Subjects: Hitch angle estimation; Kalman filter; Point set registration; Radar tracker; Track-before-detect; Trailer rotation; Track-before-detect; Hitch angle estimation; Point set registration; Trailer rotation; Kalman filter; Radar tracker
• ISSN: 0165-1684; 1872-7557
• DOI: 10.1016/j.sigpro.2021.108221

•Trailer or hitch angle can be estimated with existing radars at the rear of a vehicle.•Radars are used because they perform well in a variety of weather conditions.•Algorithm includes a trailer tracker and a procedure for point set registration.•The algorithm is tested on radar data from an experimental test bed.•Performance analysis suggests that the approach has a potential for deployment. [Display omitted] Algorithms for trailer control and backup need to keep track of the trailer angle, therefore, the angle needs to be determined. This work shows that the angle can be estimated using 2D point clouds collected from two automotive radars installed in the taillight fixtures of a trailer-coupled truck. The detection threshold of each radar is reduced to allow more detections of the trailer. A tracking procedure is introduced to find the trailer detections as the trailer rotates; the tracked set of detections are then compared with one or more reference sets to estimate the angle. Estimated angles are further refined with a Kalman filter to obtain smooth estimates of the angle. The algorithm is tested on 15 datasets; each dataset is obtained in an experiment in which the trailer is rotated up to about 40∘ in both radar directions. The estimates from all the datasets result in a global root mean squared error of about 0.9∘ for up to 10∘ absolute trailer rotation; about 1.3∘ for up to 20∘ absolute rotation; about 1.5∘ for up to 30∘ absolute rotation; and about 1.9∘ for all trailer angles considered. The algorithm executes in about 3 ms on a typical personal computer.