Three-Dimensional Object Recognition TRIZ Cases
- Object Configuration Recognition Using Orientation-Normalized Images
Orientation-normalized ROI images and histogram features classify object configuration states accurately under size and pose changes with low computing demand.
- Multi-View 3D Object Detection With Viewpoint Equivariance
View-dependent queries and equivariance loss improve multi-view 3D object localization consistency across camera angles.
- 3D Face Shape Prediction Using Synthetic Training Data
Synthetic 3D head scans train a deep model for accurate facial reconstruction with lower lighting sensitivity and less real data.
- Image Feature Extraction Using Overlap-Aware DNN Training
Automatic overlap-point labeling splits images into overlap and non-overlap regions, improving DNN feature extraction accuracy with less manual data prep.
- Point Cloud Segmentation Using Spatially Ordered 1D Convolution
Spatially ordering point clouds enables 1D convolution for semantic segmentation, cutting memory and compute while preserving local structure.
- 3D Point Cloud Segmentation Training with Image Alignment Constraints
Image-to-point-cloud alignment loss improves weakly supervised pseudo-label accuracy and raises 3D semantic segmentation precision.
- Utility Pole Photogrammetry for Electric Grid Asset Inference
Oblique aerial images and photogrammetry infer pole asset characteristics from multiple views, improving grid model coverage and update speed.
- Perception Perturbation Detection Using Cross-Model Consistency
Cross-model consistency scoring detects and localizes sensor-data perturbations, improving perception accuracy while limiting false alarms.
- Neural Event-Camera Representation for Sparse Voxel Processing
Sparse event streams are converted into trainable 3D voxel representations to improve processing efficiency, preserve features, and reduce model size.
- Neural Event-Camera Representation Using 3D Voxel Grids
Sparse event streams are converted into 3D voxel grids and neural representations to improve processing efficiency while preserving features and bandwidth.
- Multi-Camera Appearance Analysis for Full-Face Evaluation Accuracy
Multiple cameras capture front and side facial regions to improve appearance evaluation accuracy when single-view imaging misses cheek details.
- Single-Camera XR Content Creation Using 2D-to-3D Object Mapping
Converts single-camera 2D video into interactive 3D XR content through object recognition, 3D model matching, and indexed navigation.
- 3D Head Capture Using Guided Stimulus and Single-Camera Views
A moving visual or acoustic cue guides head rotation so one camera can capture multi-angle facial data for accurate 3D models with lower complexity.
- Mining Vehicle Sensor Fusion for Sparse Obstacle Detection
By aligning past and current mine-environment measurements in one coordinate frame, this case improves sparse obstacle detection for mobile mining vehicles.
- Image Normalization for Low-Power Object State Classification
Affine ROI normalization and histogram features distinguish object states despite arbitrary orientation and scale with low computing power.