Vehicle Sensor Object Detection With Linear-Scaling Prior Boxes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional object detection systems in autonomous and semiautonomous vehicles face inefficiencies due to the 'combinatorial explosion' of complexity when using multiple prior boxes, limiting the analysis of object features like orientation and velocity, and increasing computational burden.
Innovation Solution
The proposed solution involves a processor that generates a feature map from sensor information, associates prior boxes with each pixel, determines confidence values for height, width, and orientation priors, and refines these values to improve object detection accuracy, allowing for a more efficient analysis by scaling linearly and enabling the addition of priors for various properties without increasing complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple prior boxes are used for object detection, then detection coverage is improved, but computational complexity increases exponentially
Solution Approach 1:
The patent segments the object detection process into two independent phases: (1) generating a single prior box with basic properties (center, size) and (2) separately determining multiple priors for different object properties (height, width, orientation, velocity). This segmentation avoids the combinatorial explosion of traditional multi-prior-box approaches while maintaining comprehensive detection coverage.
Solution Approach 2:
The patent transitions from a two-dimensional prior box approach (width and height only) to a multi-dimensional framework that independently processes multiple property dimensions (height, width, orientation, velocity). Each dimension is handled separately through independent neural network branches, allowing linear scaling rather than exponential complexity growth.
2Adaptability or versatility
If multiple prior boxes with multiple properties are analyzed, then feature analysis capability is improved, but computational burden increases
Solution Approach 1:
The computational workload is segmented into independent processing branches for each property type (height prior, width prior, orientation prior, velocity prior). Each branch processes its specific property independently, avoiding redundant computations across all property combinations and enabling efficient parallel processing.
Solution Approach 2:
The patent performs preliminary action by generating a single prior box with basic spatial properties first, then uses this as a foundation for subsequent independent property analysis. This preliminary spatial localization reduces the search space for subsequent property detection, lowering overall computational burden.
3Reliability
If conventional object detection systems use multiple prior boxes, then object detection coverage is improved, but system efficiency decreases
Solution Approach 1:
The detection system is segmented into independent modular components that process different property types separately. This modular architecture enables parallel execution of height, width, orientation, and velocity detection, significantly improving processing efficiency while maintaining comprehensive detection coverage through the aggregation of results from all modules.
Data Source
AI summary
Methods of processing vehicle sensor information for object detection may include capturing generating a feature map based on captured sensor information, associating with each pixel of the feature map a prior box having a set of two or more width priors and a set of two or more height priors, determining a confidence value of each height prior and each width prior, outputting an indication of a detected object based on a highest confidence height prior and a highest confidence width prior, and performing a vehicle operation based on the output indication of a detected object. Embodiments may include determining for each pixel of the feature map one or more prior boxes having a center value, a size value, and a set of orientation priors, determining a confidence value for each orientation prior, and outputting an indication of the orientation of a detected object based on the highest confidence orientation.