LiDAR Millimeter Wave Radar Feature Integration for Object Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing vehicle sensor systems struggle to improve object recognition accuracy by simply integrating sparse and dense sensing results from multiple sensors, leading to contradictory recognition results.
Innovation Solution
An information processing device and method that acquire feature amounts from both sparse and dense sensing results and calculate integrated feature amounts based on relative distances between detection points, using a configuration that includes an integration processing unit, restoration unit, and semantic segmentation unit to enhance object recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If sensing results from multiple sensors are simply integrated, then processing is simplified, but object recognition accuracy deteriorates due to contradictory results between sparse and dense sensing data
Solution Approach 1:
The patent transforms the integration approach by changing the parameter of how feature amounts are combined. Instead of simple integration, it calculates integrated feature amounts based on relative distances between detection points, weighting the contribution of each sensor's data according to its spatial relationship with the object. This resolves the contradiction by maintaining processing simplicity while improving recognition accuracy through distance-based parameter adjustment.
Solution Approach 2:
The patent introduces an intermediary calculation step that processes the feature amounts from multiple sensors before final integration. The integrated feature amount calculation unit acts as a mediator that reconciles contradictory results by considering relative distances between detection points, thereby improving accuracy without significantly complicating the overall processing flow.
2Productivity
If feature amounts from multiple sensors are integrated without considering relative distances, then processing is faster, but recognition accuracy deteriorates due to contradictory sensing results
Solution Approach 1:
The patent modifies the integration parameter by incorporating relative distance information into the feature amount calculation. This allows the system to process data efficiently while improving accuracy, as the distance-based weighting can be computed quickly and resolves contradictions between sparse and dense sensing results without requiring complex processing.
3Measurement precision
If dense sensing data from LiDAR or stereo cameras is used alone, then detailed three-dimensional point clouds are obtained, but processing complexity increases and sparse detection capability is lost
Solution Approach 1:
The patent merges dense sensing data from LiDAR/stereo cameras with sparse sensing data from other sensors by integrating their feature amounts based on relative distances. This combination maintains the detailed three-dimensional point cloud information while reducing processing complexity through the unified integration approach, and also preserves sparse detection capabilities by incorporating multiple sensor types.
4Device complexity
If sparse sensing data from millimeter wave radar is used alone, then processing is simpler, but detection detail and recognition accuracy deteriorate
Solution Approach 1:
The patent combines sparse sensing data from millimeter wave radar with dense sensing data from other sensors. The integration unit merges these different types of data by calculating integrated feature amounts based on relative distances, thereby maintaining processing simplicity while enhancing detection detail and recognition accuracy through the complementary information from dense sensors.
Data Source
AI summary
The present disclosure relates to an information processing device, an information processing method, and a program for improving object recognition accuracy.A feature amount obtained by a point cloud that is a dense detection point of LiDAR or the like and a feature amount of a sparse detection point of a millimeter wave radar or the like within a predetermined distance are grouped and integrated with reference to the position of the feature amount of the dense detection point, the point cloud is restored using an integrated feature amount of the dense detection point, and object recognition processing such as semantic segmentation is performed. The present disclosure can be applied to a mobile body.