Path Prediction Model Training for Dynamic Vehicle Appearance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies for training models to predict a vehicle's driving path fail to accurately account for dynamic objects disappearing from or appearing within the camera's field of view, leading to low accuracy in real-world environments.
Innovation Solution
An apparatus and method that utilize a sensor to obtain a time series of training images in a real environment and a controller to train a path prediction model, recognizing dynamic objects that disappear or appear by establishing training strategies based on current and future time points, using a transformer network to predict the location of vehicles with feature information such as location, speed, and heading angle.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing technology trains the prediction model based on a fixed number of vehicles in training images, then the model structure is simple and training is straightforward, but the prediction accuracy is low in real road environments where vehicles dynamically appear and disappear
Solution Approach 1:
The patent applies dynamics by making the number of vehicles in training images variable rather than fixed. The training method dynamically adjusts the number of vehicles based on real road environment conditions, allowing the model to learn from scenarios where vehicles appear and disappear. This dynamic approach resolves the contradiction by enabling high prediction accuracy in realistic conditions while managing training complexity through systematic data processing procedures.
Solution Approach 2:
The patent changes the parameter of vehicle count from a fixed value to a variable parameter that reflects real-world conditions. By varying the number of vehicles in training images according to actual road scenarios, the model learns to handle dynamic environments effectively. This parameter change enables the model to achieve high prediction accuracy while the training complexity is managed through structured data preparation and processing methods.
2Reliability
If the model is trained to recognize all vehicles in every training image, then complete path prediction is possible, but the model fails when vehicles disappear or appear in the field of view
Solution Approach 1:
The patent makes the training data dynamic by allowing vehicles to appear and disappear in training images, mirroring real road conditions. This dynamic training approach enables the model to adapt to changing environments while maintaining reliable performance. The model learns to handle variable vehicle presence, resolving the contradiction between reliability and adaptability.
Solution Approach 2:
The patent incorporates feedback mechanisms in the training process by using multiple training images with varying vehicle configurations. The model receives feedback from diverse training scenarios where vehicles dynamically appear and disappear, enabling it to adapt to real-world conditions while maintaining reliable prediction performance across different situations.
3Productivity
If training data is processed to include only fixed-number vehicle scenarios, then data processing is simple and fast, but the model cannot handle real-world dynamic road environments
Solution Approach 1:
The patent applies preliminary action by pre-processing training images to include varied vehicle configurations that reflect real road conditions. Data processing is performed in advance to create a comprehensive training dataset with dynamic vehicle scenarios. This preliminary preparation enables efficient training while achieving high prediction accuracy, as the model is pre-exposed to diverse situations it will encounter in deployment.
Solution Approach 2:
The patent maintains continuity of useful action by systematically processing training data to continuously include dynamic vehicle scenarios. The training process continuously exposes the model to varied vehicle configurations, ensuring that the model learns to handle real-world conditions. This continuous exposure maintains training efficiency while progressively improving prediction accuracy through comprehensive data coverage.
Data Source
AI summary
An apparatus for training a driving path prediction model of a vehicle and a method therefor are disclosed. The apparatus includes a sensor that obtains a time series of training images in a real environment and a controller that trains a path prediction model based on dynamic objects in the time series of training images. The controller is configured to train the path prediction model to recognize a dynamic object disappearing from the time series of training images and a dynamic object appearing in the time series of training images.