Vehicle Surroundings Representation With Temporal Feedback Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional tracking-by-detection paradigms in advanced driver assistance systems and autonomous driving face challenges in efficiently utilizing sensor inputs over time, particularly in dynamic environments, leading to suboptimal performance in tasks like object detection, semantic segmentation, and drivable space identification.
Innovation Solution
A method utilizing a machine learning model with a preprocessing module and task-specific modules that incorporates temporal feedback and physical models to process sensor data, enabling simultaneous performance of multiple tasks and leveraging temporal correlations to enhance accuracy and robustness.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional tracking-by-detection paradigm is used with separate detector and tracking algorithms, then the system structure is simple and easy to implement, but the performance in dynamic environments is suboptimal and computational efficiency is reduced
Solution Approach 1:
The patent merges the detector algorithm and tracking algorithm into a single integrated neural network model. The detector modules process sensor inputs to generate detection results, while the tracker modules simultaneously perform tracking by utilizing temporal correlations from historical detection data. This unified architecture eliminates the need for separate processing stages, improving performance in dynamic environments while maintaining computational efficiency through shared feature extraction and coordinated optimization.
2Productivity
If multiple separate algorithms are used for different tasks (object detection, semantic segmentation, drivable space identification), then each task can be optimized independently, but the computational effort and training data requirements increase significantly
Solution Approach 1:
The patent implements a multi-functional neural network architecture where a single model performs multiple tasks including object detection, semantic segmentation, and drivable space identification. The detector modules generate task-specific outputs for each function, while the tracker modules collectively handle temporal tracking across all tasks. This universal architecture shares computational resources and learning parameters across tasks, reducing training data requirements and improving computational efficiency compared to multiple separate algorithms.
3Measurement precision
If temporal feedback is not incorporated in the processing of sensor inputs over time, then the processing is simpler and faster for single measurements, but the accuracy in dynamic environments deteriorates
Solution Approach 1:
The patent incorporates temporal feedback mechanisms where the neural network processes sensor inputs sequentially over time. The tracker modules receive detection results from previous time steps and use this historical information to improve current detection accuracy. This feedback loop allows the system to leverage temporal correlations and motion patterns, enhancing measurement precision in dynamic environments while managing processing complexity through efficient recurrent processing.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The invention relates to a method (100) for determining an environment representation of the environment of a vehicle (1), comprising the following steps: - providing (101) input data, wherein the input data comprises sensor data (3) and feedback data (4), wherein the sensor data (3) results from a detection by at least one sensor (2) of the vehicle (1), and wherein the sensor data (3) represents a detection of the environment of the vehicle (1), - providing (102) a machine learning model (9), wherein the machine learning model (9) comprises a preprocessing module (5) and at least one task-specific module (6), - providing (103) the feedback data (4), wherein the feedback data (4) comprises at least one past output (7) of the at least one task-specific module (6) and/or at least one past output of the preprocessing module (5).wherein the past output (7) was determined at least one iteration prior to a current iteration by the at least one task-specific module (6) and/or the preprocessing module (5), - extracting (104) features from the input data by the preprocessing module (5), - determining (105) a respective output (7) by the at least one task-specific module (6) based on the features extracted by the preprocessing module (5) and/or the at least one past output (7) of the at least one task-specific module (6) and/or the at least one past output of the preprocessing module (5) for the current iteration in order to determine the environment representation of the environment of the vehicle (1). The invention further relates to a computer program, a device, and a storage medium for this purpose.