Adaptive BEV Grid Partitioning for Multicamera Fusion Alignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous driving systems face challenges in accurately detecting and tracking objects due to the mismatch between fisheye and long-range cameras, leading to spatial misalignment and inefficient use of computational resources in Bird's Eye View (BEV) grids.
Innovation Solution
Adaptive grid partitioning in the BEV space, where grid sizes are adjusted based on the camera type and detection range, allocating higher resolution to areas with high-resolution fisheye coverage and coarser grids to regions with lower resolution from long-range cameras.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a uniform BEV grid is used for all areas, then the grid structure is simple and easy to implement, but it creates redundancy in overlapping areas and requires handling distortion for accurate object representation
Solution Approach 1:
The patent applies local quality by using different grid resolutions in different regions of the BEV space. High-resolution grids are used in areas covered by fisheye cameras where detailed object representation is critical, while coarser grids are used in areas covered by long-range cameras where the detection range is limited. This allows the system to maintain simple grid structure in most areas while providing high precision where needed.
Solution Approach 2:
The patent implements dynamics by making the grid resolution adaptive rather than fixed. The grid cell size is dynamically adjusted based on the camera type and detection range for each region. This dynamic adaptation allows the system to optimize between computational efficiency and measurement precision according to the specific requirements of each BEV region.
2Measurement precision
If high-resolution grids are used for all areas to maintain detection precision, then object representation accuracy is improved, but computational resources are wasted in areas where high resolution is not needed
Solution Approach 1:
The patent applies local quality by allocating high-resolution grids only to regions where fisheye cameras provide coverage and detailed object representation is necessary. In regions covered by long-range cameras with limited detection range, the system uses coarser grids that require less computational power. This localized application of high resolution optimizes the balance between detection precision and computational resource consumption.
Solution Approach 2:
The patent implements partial action by applying high-resolution processing only to the extent necessary for accurate detection in fisheye coverage areas, rather than uniformly across the entire BEV space. This partial application of high resolution avoids excessive computational resource usage in regions where the coarser grid from long-range cameras is sufficient.
3Area of stationary object
If fisheye cameras with wider FOV are used to capture more area, then coverage area is improved, but significant overlap between images creates redundancy and requires handling distortion
Solution Approach 1:
The patent applies local quality by using coarser grid resolution in overlapping regions covered by both fisheye and long-range cameras. This reduces the quantity of data needed to represent these overlapping areas, effectively handling the redundancy issue. The finer grid is used only in non-overlapping fisheye regions where detailed representation is critical.
Data Source
AI summary
A method for generating an adaptive Birds-Eye-View (BEV) grid includes obtaining sensor data generated by one or more sensors of a vehicle, wherein the sensor data includes one or more images captured by one or more cameras of a first type having a first detection range and one or more images captured by one or more cameras of a second type having a second detection range; extracting, from the sensor data, a plurality of features to generate a plurality of multi-scale image features; projecting the plurality of multi-scale image features onto a BEV space representing an environment surrounding the vehicle; generating an adaptive BEV grid comprising a plurality of grid cells that incorporates a combination of the plurality of multi-scale image features; and adjusting a size of one or more of the plurality of grid cells based on one or more pre-defined factors.