Scene Recognition TRIZ Cases

A vehicle detection system calculates turning-off probability using sensor data to adjust adaptive cruise control parameters.

A controller uses a recurrent neural network to predict interactive motions of surrounding vehicles for lane change decisions.

Evaluation unit adds 3D object shapes to the projection surface of a vehicle surround view system.

A Bayesian object classifier system processes multi-sensor outputs to generate confidence score distributions for autonomous vehicle control.

A convolutional neural network classifies digital image representations of vehicle telemetry data to identify traffic control features.

Matching visual features against reference data prevents unauthorized rerouting and theft of autonomous vehicles.

An actor importance model ranks detected objects using neural embeddings to direct autonomous vehicle processing resources.

A vehicle control apparatus selects nearby lane division lines to generate a target path for accurate travel.

A particle-based hazard detection system fuses multi-sensor outputs to generate confidence-weighted location estimates.

Simulated spectral representation guides hyperspectral sensor band selection, reducing processing loads and communication bandwidth requirements.

A generation unit projects three-dimensional point clouds onto a two-dimensional plane to create sensing images with pixel values corresponding to speed information.

Controller analyzes sensor data to assess narrow road passage feasibility, enabling safe autonomous navigation for novice drivers.

Automated image processing system captures visual data and retrieves targeted prompts based on recognition results.

Aerial photography captures surface illumination to generate pixel-level lumen data, replacing slow ground measurements with zone-based threshold assessment.

Multi-task learning merges detection and recognition into one model, reducing processing time while maintaining accuracy.

A color reproducing unit selects the highest occupancy Bayer phase to generate luminance values for block matching.

Analog CEM circuit replaces digital signal processing to reduce power consumption while maintaining target detection precision.

Training a substitute model to replicate black box object detection outputs enables effective adversarial attacks without internal algorithm details.

A classification and segmentation system processes SAR image pixels to group terrain types.

Thermal radiation analysis filters non-navigable obstacles to predict accurate friction coefficients, preventing abrupt safety interventions.

A cloud-based analytics platform processes video inputs from unmanned aerial vehicles to generate three-dimensional surveillance data.

A permutation invariant convolution layer segments video streams and identifies high-likelihood frames to generate global representations.

A computing device selects training examples based on derivative vector norms and importance scores to train machine learning models.

Target projectors display visual markers along automated guided vehicle travel paths, resolving unclear robot intentions that risk human obstruction.

Continuous speed constraint functions integrate path curvatures and obstacles to resolve optimization complexity in autonomous vehicle navigation.

Automated digital elevation model validation adjusts z-values based on terrain slope to ensure data consistency without human intervention.

Dual lenses with distinct angles of view acquire parallax image data, resolving the trade-off between wide monitoring coverage and precise distance measurement.

A mobile robot adjusts its distance and direction using predicted user location data.

Sequential CNN and RNN models resolve lighting misclassification errors in dynamic environments.