Neural Network Map Deviation Detection for Automated Vehicles
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Automated vehicles rely on high-definition map data, but this data can become outdated or incorrect due to environmental changes, leading to map deviations that may pose safety risks if undetected, especially in conditions like bad weather or construction sites, where existing detection methods perform poorly.
Innovation Solution
A computer-implemented method using a neural network that encodes both map and sensor data as feature maps, allowing for the detection and correction of map deviations by comparing real-world sensor data to stored map data, with the network predicting verification, deletion, insertion, or substitution of map elements, and utilizing specialized anchor grids and 3D data processing to enhance performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing detection methods are used for map elements, then detection can be performed, but performance decreases in bad weather conditions or when elements are occluded
Solution Approach 1:
The patent combines multiple data sources (sensor data, map data, and weather condition data) into a unified detection system. The neural network integrates these diverse inputs to compensate for the weaknesses of individual sensors in adverse conditions, thereby maintaining reliable detection performance when elements are occluded or weather is poor.
Solution Approach 2:
The patent introduces an intermediary processing layer (the neural network with multiple input channels) that mediates between raw sensor data and final detection results. This intermediary structure allows the system to process and reconcile conflicting or degraded information from different sources, improving robustness against weather conditions and occlusion.
2Reliability
If map data is updated frequently to maintain accuracy, then reliability improves, but time consumption and processing complexity increase
Solution Approach 1:
The patent performs preliminary processing of sensor data and map data into feature maps before comparison. By pre-encoding the data into standardized feature representations, the system reduces the computational burden during real-time deviation detection, enabling faster update cycles without sacrificing accuracy.
Solution Approach 2:
The patent replaces traditional mechanical comparison methods with a neural network-based approach. The neural network efficiently compares feature maps and detects deviations through learned patterns, significantly reducing processing time compared to conventional algorithms while maintaining high detection accuracy.
3Measurement precision
If detailed comparison methods are used to detect deviations, then measurement precision improves, but device complexity increases
Solution Approach 1:
The patent transforms the comparison task from operating on raw sensor data to operating on encoded feature maps. By changing the parameter space from pixel-level or point-cloud-level data to abstracted feature representations, the system achieves high measurement precision while reducing computational complexity through the encoding process.
Data Source
Figure 1
Figure 2~3
Figure 4
AI summary
The invention refers to a method for deviation detection, system, and vehicle. The method for a detection of map deviations comprising encoding map data (55) and receiving a first feature map (57), encoding sensor data (58) and receiving a second feature map (60), wherein the sensor data is representing one or a plurality of real world objects in a vehicle environment. The method further comprising providing the first feature map (57) and the second feature map (60) as input data to a neural network (50, 53) and detecting deviations between the sensor data (55) and the map data (58).