Vehicle Position Correction Using RNN-Based Temporal Driving Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current deep learning techniques, such as convolutional neural networks, do not effectively consider the temporal order of input data, making it difficult to accurately correct the position of a vehicle based on its driving information.
Innovation Solution
A vehicle position correction apparatus and method that utilizes deep learning to predict the position of a probe vehicle based on its driving information, including position, speed, acceleration, and road section data, and corrects the actual position of a target vehicle using a completed model, considering the temporal aspects through a recurrent neural network structure.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a convolutional neural network is used for position prediction, then the model can process input data, but it cannot effectively consider the temporal order of driving information
Solution Approach 1:
The patent changes the architectural parameters of the neural network from a standard CNN to a hybrid structure incorporating recurrent neural network units (LSTM or GRU). This structural parameter change enables the model to process sequential driving information while preserving temporal order, thereby improving position prediction accuracy without losing temporal information.
2Device complexity
If GPS measurement is used alone, then the system is simple, but the position accuracy deteriorates in areas with poor satellite reception such as underpasses and branch roads
Solution Approach 1:
The patent creates a composite positioning system that combines GPS satellite positioning with deep learning-based position prediction. This composite approach uses the strengths of both methods: GPS provides global coverage when available, while the deep learning model provides accurate predictions in GPS-denied areas, achieving high overall accuracy without excessive complexity.
Solution Approach 2:
The deep learning model acts as an intermediary between GPS measurements and the final position output. When GPS signals are available, the model processes them normally; when GPS signals are poor or unavailable, the model uses learned patterns from driving information to predict position, effectively mediating the positioning process to maintain accuracy across all conditions.
Data Source
AI summary
A vehicle position correction apparatus and a method thereof may include a learner that deep learns a model which predicts a position of a probe vehicle based on driving information of the probe vehicle traveling on a road, a communication device that receives driving information of a target vehicle from the target vehicle, and a controller that obtains a predicted position of the target vehicle based on the model on which the deep learning is completed and corrects an actually measured position of the target vehicle to the predicted position of the target vehicle.