Occupancy Masking for Grid-Based Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for processing measurement data from area surveillance into classification scores and object properties often result in 'ghost' objects appearing in unoccupied cells, leading to inappropriate reactions in applications like vehicle trajectory planning.
Innovation Solution
The method involves projecting point cloud data onto a grid, using a task neural network to process the grid into an output grid with classification scores and object properties, and applying 'occupancy masking' to unoccupied cells by setting their values to neutral, ensuring accurate object detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If measurement data is projected onto a regular grid for processing, then the data can be systematically analyzed to determine object properties, but 'ghost' objects appear in unoccupied cells leading to false detections
Solution Approach 1:
The patent applies occupancy masking before the neural network processes the grid data. By pre-marking unoccupied cells with neutral values based on point cloud occupancy information, the system prevents false detections from occurring in the first place, rather than trying to correct them afterward
Solution Approach 2:
The patent applies different treatments to different regions of the grid: occupied cells retain their measurement data for normal processing, while unoccupied cells are masked with neutral values. This localized differentiation ensures that false detections are prevented only where necessary without affecting the processing of valid object regions
2Productivity
If the neural network processes all grid cells including unoccupied ones, then comprehensive coverage is achieved, but computational complexity increases and detection performance decreases
Solution Approach 1:
The patent extracts and removes unoccupied cells from the processing pipeline by masking them with neutral values before neural network processing. This eliminates the need for the network to analyze irrelevant empty regions, reducing computational complexity and improving detection performance on actual objects
Solution Approach 2:
Instead of processing all grid cells uniformly, the patent applies partial processing by selectively masking unoccupied cells. This partial action approach focuses computational resources only on occupied cells that contain actual objects, improving overall processing efficiency
Data Source
AI summary
A method for processing measurement data from a surveillance of an area into classification scores with respect to a given classification and/or properties of objects in the area. The method includes: providing a point cloud of measurement data that assigns, to each of a plurality of points in space, measurement values of at least one quantity; providing an input grid with a plurality of cells; assigning, based on the point cloud, values of the at least one measurement quantity, and/or of at least one work product derived therefrom, to each cell of the input grid; and processing, by a task neural network, the input grid into an output grid whose cells carry classification scores and/or properties of objects as values.