Trailer Angle Estimation via ML Keypoint Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional techniques for estimating the trailer angle are susceptible to lighting and environmental changes, leading to inaccurate and failed estimates, and are not suitable for real-time deployment due to increased latency.
Innovation Solution
The use of machine learning models to predict keypoints on the center axis of the trailer draw bar, back-project these keypoints onto a 3D representation of the ground, and calculate the trailer angle, allowing for estimates independent of draw bar design and robust across varying lighting conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional feature extraction techniques are used to identify trailer features, then the system can track keypoints from frame to frame, but the estimation becomes highly susceptible to lighting and environmental changes resulting in inaccurate estimates
Solution Approach 1:
The patent replaces conventional mechanical feature extraction methods (edge detection, contrast-based tracking) with a machine learning model that learns robust visual features. The ML model processes the entire image to predict keypoint locations directly, eliminating reliance on contrast-based edge detection that fails under varying lighting conditions. This substitution enables reliable trailer angle estimation across diverse environmental conditions.
2Measurement precision
If conventional edge extraction and line fitting techniques are used, then the system can calculate trailer angle, but the computational resources required increase latency making it unsuitable for real-time deployment
Solution Approach 1:
The patent segments the trailer angle estimation problem into two efficient components: (1) an ML model that quickly predicts keypoint locations from the image, and (2) a simplified geometric calculation that computes the angle from these keypoints. This segmentation avoids the computationally intensive iterative optimization of conventional line-fitting methods while maintaining measurement precision.
Solution Approach 2:
The patent substitutes conventional computational geometry methods (edge detection, Hough transform, iterative line fitting) with a machine learning model that directly regresses keypoint coordinates. This substitution dramatically reduces computational complexity and processing time, enabling real-time deployment while preserving measurement accuracy through the ML model's learned feature representations.
3Ease of operation
If contrast-based feature tracking is used to identify trailer components, then the system can determine trailer angle, but object tracking fails in situations where relative contrast is not detectable such as under shadows
Solution Approach 1:
The patent trains the machine learning model on diverse training data that includes various lighting conditions, shadows, and environmental scenarios. This enables the model to universally recognize trailer features across all conditions, not just in ideal contrast-based scenarios. The model learns invariant features that remain detectable regardless of lighting variations, ensuring reliable tracking functionality under shadows and other challenging conditions.
Data Source
AI summary
In various examples, a trailer angle may be estimated using one or more machine learning models to predict one or more keypoints on the center axis of the trailer drawbar (e.g., a keypoint representing the drawbar junction around which the drawbar pivots, one or more other keypoints along the center axis), back-projecting the predicted keypoint(s) onto a three-dimensional (3D) representation of the ground, and calculating the angle between the longitudinal axis of the towing vehicle and a line or ray formed by or fitted to the projected keypoints. The trailer angle may be estimated at any frame rate. For each frame, keypoints may be predicted from that frame and/or optical flow or some other type of feature tracking may be used to propagate predicted keypoint(s) from a preceding frame in lieu of predicting keypoint(s), and the resulting keypoint(s) may be used to estimate the trailer angle for that frame.