Map Building Using Heat Map AI Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for building high-definition maps, particularly global high-definition maps, suffer from low robustness and usability due to the repetitive merging of local curves, which can lead to incorrect merging results and require excessive threshold value settings.
Innovation Solution
A method involving the generation of a heat map based on first curves derived from input frames, followed by the use of an AI network to obtain second curves, which are then used to identify map elements. This process transforms the problem of building a global map into a collective prediction problem, enhancing robustness and versatility.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If local curves are repeatedly merged using incremental serialization to build global maps, then map coverage area increases, but map building robustness deteriorates
Solution Approach 1:
The patent segments the global map building process into multiple local map building tasks. Each local map is built independently using vectorized map annotation (VMA) on individual frames or frame sequences, avoiding the accumulation of errors that occurs when repeatedly merging local curves. The segmented local maps are then integrated to form the complete global map, maintaining both coverage area and robustness.
Solution Approach 2:
The patent introduces an intermediary representation layer (local map elements such as lanes, intersections, and road boundaries) that mediates between raw curve data and the final global map. Instead of directly merging curves, the system converts curves into structured local map elements first, which then serve as the basis for global map construction. This intermediary step eliminates the need for excessive threshold value settings and improves merging accuracy.
2Ease of manufacture
If incremental serialization is used to merge vectorized map annotations, then global map can be constructed from local maps, but usability deteriorates due to low robustness
Solution Approach 1:
The patent performs preliminary actions by pre-processing input frames to extract high-quality local map elements before global map construction. Local maps are built with complete vectorized annotations including lanes, intersections, and road boundaries in advance, ensuring that all necessary map elements are properly identified and structured before the global integration phase. This preliminary preparation eliminates the need for repeated curve merging and threshold adjustments during global map construction.
Solution Approach 2:
The patent changes the fundamental parameters of map representation from curve-based incremental serialization to element-based structured annotation. By transitioning from merging curves to integrating pre-annotated map elements (lanes, intersections, road boundaries), the system achieves both global map construction capability and improved usability. The parameter change from continuous curve data to discrete map elements eliminates the robustness and usability issues associated with traditional incremental serialization.
Data Source
AI summary
Disclosed are a map-building method, a map-building apparatus, an electronic device, and a non-transitory computer-readable storage medium, which relate to the field of artificial intelligence (AI). The map-building method includes generating a heat map based on first curves used to identify a map element of at least one local area and obtaining second curves using an AI network, based on the heat map, in which the second curves are used to identify a map element of the map, and the map-building method transforms a global map-building problem into a collective prediction problem and performs prediction based on image detection using the heat map as an input so that various cases are handled in an integrated manner, and the excessive use of a threshold value is avoided, thereby having excellent versatility.