3D Model Projection for Vehicle Classification in Aerial Imagery
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current 2D and 3D object classification methods are inadequate for distinguishing between types of vehicles in aerial imagery due to limitations in leveraging 3D shapes, handling view variance, and accurately simulating scene conditions, especially when faced with clutter, noise, and fine-level distinctions.
Innovation Solution
A computer-implemented method that projects 3D vehicle models with salient feature locations onto detected vehicles in aerial images, using Histogram of Oriented Gradients (HoG) feature descriptors and support vector machines (SVM) to calculate positive and negative match scores, and classify vehicles based on these scores, while also detecting vehicle parts and determining vehicle pose using multi-class classifiers and deformable models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If 2D methods are used for vehicle classification, then the system can operate with simpler processing, but the system cannot leverage 3D shape properties and fails to achieve fine-level distinction between vehicle types
Solution Approach 1:
The patent transitions from 2D image analysis to 3D model-based recognition by projecting 3D vehicle models onto 2D aerial imagery. This dimensional enhancement allows the system to leverage 3D shape properties, surface normals, and geometric relationships while still operating on 2D input images, thereby achieving fine-level vehicle type distinction without completely abandoning 2D processing advantages
Solution Approach 2:
The patent introduces 3D vehicle models as an intermediary between the 2D aerial imagery and the classification decision. These models serve as a bridge that encodes 3D shape knowledge and can be projected to match against 2D image features, enabling the system to benefit from both 2D input simplicity and 3D shape discrimination power
2Measurement precision
If prior art 3D model based recognition methods are used, then the system can utilize 3D shapes, but the system is unable to accurately simulate scene conditions and handle clutter, noise, and blurry imagery
Solution Approach 1:
The patent adjusts key parameters in the rendering pipeline including illumination models, camera projection parameters, and texture mapping to match actual aerial imaging conditions. By changing these parameters to reflect real-world scene conditions such as aerial viewpoint geometry and lighting, the rendered models become more comparable to actual imagery, improving reliability in cluttered and noisy environments
Solution Approach 2:
The patent creates rendered copies of 3D vehicle models that simulate the appearance and characteristics of actual aerial imagery. These rendered copies include simulated camera projections, lighting effects, and potential degradation effects that mirror real scene conditions, allowing the system to compare against realistic representations rather than idealized 3D models
3Adaptability or versatility
If 2D methods with multiple single-view detectors are used, then the system can handle view variance, but the system requires arbitration logic and multiple independent detectors increasing system complexity
Solution Approach 1:
The patent merges multiple view-dependent detection capabilities into a single unified 3D model-based detector. Instead of maintaining separate 2D detectors for different views, the system uses a single 3D model that can be projected and compared against the image regardless of viewpoint, eliminating the need for arbitration logic while maintaining view variance handling through the inherent 3D geometry
Data Source
AI summary
A computer implemented method for determining a vehicle type of a vehicle detected in an image is disclosed. An image having a detected vehicle is received. A number of vehicle models having salient feature points is projected on the detected vehicle. A first set of features derived from each of the salient feature locations of the vehicle models is compared to a second set of features derived from corresponding salient feature locations of the detected vehicle to form a set of positive match scores (p-scores) and a set of negative match scores (n-scores). The detected vehicle is classified as one of the vehicle models based at least in part on the set of p-scores and the set of n-scores.