Camera-LiDAR Fused Object Detection with Adaptive Point Pruning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Real-time prediction of actions by drivers and pedestrians is challenging for semi-autonomous or autonomous vehicles, especially when they break traffic rules, leading to unsafe vehicle behavior due to either overly cautious or optimistic predictions.
Innovation Solution
The implementation of Camera-LiDAR Fused (CLF) object detection with point pruning, which involves generating a pruned LiDAR dataset by downsampling points based on planned vehicle trajectories, point labels, distances, and image data to enhance object detection and collision avoidance capabilities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the LiDAR dataset is densely sampled to improve object detection accuracy, then the detection precision improves, but the computational load and processing time increase
Solution Approach 1:
The patent applies different downsampling rates to different spatial regions based on their importance to the vehicle. Regions closer to the vehicle or containing detected objects are downsampled at lower rates (retaining more points), while distant regions are downsampled at higher rates. This local differentiation maintains detection accuracy in critical areas while reducing overall computational load.
Solution Approach 2:
The patent dynamically adjusts the downsampling parameter (downsampling rate) based on point characteristics such as distance from the vehicle, point density, and detected object locations. By changing this parameter adaptively across different regions, the system optimizes the balance between detection accuracy and processing efficiency.
2Reliability
If the LiDAR dataset is densely sampled to improve collision risk assessment, then the reliability of safety assessment improves, but the energy consumption increases
Solution Approach 1:
The system applies higher sampling density (lower downsampling rate) to regions that are more critical for collision risk assessment, such as areas closer to the vehicle or containing detected objects. Less critical distant regions are processed with higher downsampling rates, reducing overall energy consumption while maintaining reliability where it matters most.
Solution Approach 2:
The patent applies downsampling selectively rather than uniformly - using aggressive downsampling only where sufficient data remains for safe assessment, and conservative downsampling in critical regions. This partial application of the downsampling technique maintains adequate reliability while reducing energy consumption compared to uniform high-density processing.
3Ease of operation
If the LiDAR dataset is pruned to reduce processing complexity, then the ease of operation improves, but the detection precision may deteriorate
Solution Approach 1:
The patent implements region-specific downsampling where the pruning intensity varies by spatial location. Critical regions (closer to vehicle, containing objects) use lower downsampling rates to preserve detection precision, while non-critical distant regions use higher rates to reduce complexity. This local differentiation resolves the contradiction between ease of processing and detection precision.
Solution Approach 2:
The system performs preliminary object detection on the full LiDAR dataset to identify regions containing objects or potential hazards. Based on these preliminary results, it then applies adaptive downsampling that preserves precision in identified critical regions while reducing complexity in non-critical areas, ensuring detection precision is maintained where needed.
Data Source
AI summary
Systems/methods for object detection. The methods comprise: obtaining, by a computing device, a LiDAR dataset generated by a LiDAR system of the autonomous vehicle; and using, by a computing device, the LiDAR dataset and at least one image to detect an object that is in proximity to the autonomous vehicle. The object is detected by: generating a pruned LiDAR dataset by reducing a total number of points contained in the LiDAR dataset; and detecting the object in a point cloud defined by the pruned LiDAR dataset. The object detection may be used by the computing device to facilitate at least one autonomous driving operation.