GPS-Labeled Hitch Angle Estimation Without Trailer Markers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing hitch angle estimation methods using deep neural networks (DNNs) are hindered by the need for sensors or markers in learning data, leading to improper estimation when these features are absent in inference images.
Innovation Solution
A hitch angle estimation method utilizing a model trained with GPS-based orientation differences between a towing vehicle and a trailer, using images from a learning camera and GPS signals to estimate hitch angles without requiring sensors or markers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If sensors or markers are included in the learning data to obtain hitch angle, then the hitch angle can be obtained for learning, but the learning becomes improper and the DNN cannot appropriately estimate the hitch angle when these features are absent in inference images
Solution Approach 1:
The patent uses GPS coordinates to create a virtual copy of the trailer's position and orientation data, replacing the need for physical sensors or markers in images. The hitch angle is calculated from the difference between the towing vehicle's orientation and the trailer's orientation derived from GPS coordinates, creating a label that reflects real-world geometry without requiring artificial features in the learning images.
Solution Approach 2:
The patent replaces the mechanical/optical measurement system (sensors or markers attached to the trailer) with a satellite-based navigation system (GPS). Instead of using cameras to detect physical markers or sensors on the trailer, the system uses GPS coordinates to calculate the trailer's orientation and derive the hitch angle, substituting a different physical measurement paradigm that doesn't require modifying the trailer with additional hardware visible in images.
2Loss of information
If sensors or markers are attached to the learning trailer to obtain hitch angle data, then the hitch angle can be measured, but the learning trailer images become improper for training the DNN model
Solution Approach 1:
The patent creates a virtual representation of the trailer's spatial configuration using GPS coordinates instead of requiring physical sensors or markers to be visible in images. The hitch angle information is copied from the GPS-derived orientation data, allowing the learning process to access accurate hitch angle measurements while keeping the learning images clean and representative of real-world conditions without artificial features.
Data Source
AI summary
A hitch angle estimation device acquires an image of a trailer shot by a camera mounted on a vehicle towing the trailer, and estimates a hitch angle of the trailer based on the image of the trailer by using a model obtained by performing learning using learning data which is a data set of an image of a learning trailer shot by a learning camera mounted on a learning vehicle which tows the learning trailer, and a label indicating the hitch angle of the learning trailer. A difference between an orientation of the learning vehicle calculated based on a GPS signal received by a GPS receiver mounted on the learning vehicle and an orientation of the learning trailer calculated based on a GPS signal received by a GPS receiver mounted on the learning trailer is used as the hitch angle of the learning trailer.