Scene Recognition TRIZ Cases

A trained neural network generates pixel offset vectors pointing to unique object centers, forming cluster distributions that define margins for accurate mask assignment.

A geographic information processing device identifies polygon gradients to automatically set terrain attributes.

Monocular electro-optical camera tracks feature points across image sequences to generate 3D cloud representations.

Technologically diversified sensor units evaluate surroundings data for plausibility to resolve detection accuracy losses from glare and shadow.

Detects dominant edges via horizontal derivative integration to measure object width, resolving classification ambiguity between vehicles and pedestrians.

Virtual road models enable long-range object detection without complex LIDAR hardware, reducing device complexity while maintaining precision.

A computer-implemented method uses a Bounding Box Sampler and Few-Shot Classification to label objects in unlabeled image data.

A rail area extraction method based on laser point cloud data uses side walls as reference objects to determine boundaries.

Independent input stages feed a shared processing stage to generalize across sensor modalities, reducing retraining time when hardware changes.

Regression analysis of multi-vehicle driving data generates dynamic maneuver thresholds that align automated control with human roadmanship norms.

An electronic device processes satellite images using overlapping segmentation to extract multi-dimensional environmental features through an association network.

A 3D data generation unit flattens deformed shipping labels to correct distortions before recognition.

Combines visual recognition with satellite positioning to resolve urban canyon location errors.

A tensor image mapping device selects weight values via threshold comparisons to replace multiplication operations.

A hierarchical classifier system detects bicyclists using synthetic training data generated from 3D models.

Navigation assistance system detects external objects and pedestrian paths outside driver line of sight to reduce manual intervention.

A spectral geographic information system translates large hyperspectral image data into a compact database structure for efficient analysis.

Processor switches rear camera feeds based on towing status, resolving insufficient coverage when vehicles tow trailers.

Integer arithmetic constructs consolidated occupancy grids from sensor data without floating-point operations.

A mobile platform system moves autonomously to detect individuals and display personalized multimedia content segments.

A processor generates a single feature map to classify objects and regions of interest simultaneously.

Maximum spanning tree losses train neural networks to cluster social groups from RGB images without requiring explicit order labels.

Ranking neurons by activation scores prunes deep neural networks, reducing computing power and memory requirements while maintaining accuracy thresholds.

A vehicle periphery monitoring system detects approaching pedestrians using a single onboard camera to calculate object size changes and image deformation.

Encoder processes captured and depth image data using a validity mask to guide convolution operations for feature extraction.

An object recognition device combines multiple feature value sets to identify targets in captured images.

Optical flow analysis differentiates curb heights from nearby objects, resolving ultrasonic sensor ambiguity in autonomous parking.

Adaptive depth of field gated imaging synchronizes pulsed illumination with sensor exposure to resolve target detection versus image blur trade-offs.

Geometric transformations and pixel modifications augment training datasets to resolve scarcity of rare occluded scenarios.

A light receiving element uses segmented photoelectric conversion units to process incident light charges across multiple phases simultaneously.

Machine learning classifiers process derived feature vectors from passive sensors to resolve platform identification ambiguity in close-proximity formations.

Grouping radar detections into consistently moving motion subsets eliminates parametric fitting errors and improves target separation accuracy.

A processing module determines necessary situation awareness levels based on driving information and adjusts stimuli to maintain user alertness.

Segmenting a spatio-temporal graph into object and scene branches transfers privileged interaction data, stabilizing performance against spurious correlations.

Directional optics resolve light angles per detector section, reducing computational complexity while maintaining measurement precision.

Manifold transfer subspace learning transforms image data into lower-dimensional subspaces using diffusion maps and TrFLDA.

A pedestrian detecting system fuses depth spatial data with image appearance features to identify targets.