Scene Recognition TRIZ Cases

A luminance obtaining unit assigns color identifiers to target portions based on brightness ranges and groups them by spatial proximity.

Training a machine-learned model to dynamically adjust sensor parameters resolves manual tuning bottlenecks and improves obstacle identification.

A Lidar processing system extracts ground-level elevation values from roadway intersections to generate accurate terrain models.

Calculating intensity slope differences between horizon regions enables threat detection systems to calibrate thresholds against visibility variations.

A vehicle-mounted measurement device unit dynamically sets overlapping detection areas between sensors to generate integrated data.

An optical marker system enables a delivery drone to identify specific recipient locations via image processing, resolving GPS coordinate limitations.

Edge device detects double parking violations by bounding road lanes and vehicles, replacing manual review with automated validation.

A vehicle exterior environment recognition apparatus pairs side and rear faces of potential three-dimensional objects using angular classification.

Derivative peaks cueing guides a convolutional neural network to classify buried objects, reducing speckle noise interference.

A neural network adjusts weighting coefficients using dual evaluation functions to align intermediate outputs from computer graphics and actual images.

A processor evaluates input data quality using a virtual model to select necessary learning data for storage.

Three-dimensional spatial analysis differentiates persons from spray or birds by verifying object origins, reducing false alarms in maritime monitoring.

A road surface measurement sensor generates signals that a controller digitalizes to calculate curvature patterns for vehicle identification.

A vehicle recognition apparatus identifies paired light sources using positional relations and derives a reliability degree for each pair.

Vehicle image processing detects road objects to determine traffic disturbance positions.

A vehicle obstacle detection device extracts precise boundary contours using a Poisson gradient vector flow algorithm for accurate type recognition.

Local reference frame encoding resolves global coordinate limitations by enabling accurate multi-agent trajectory prediction without fixed reference agents.

Image-based rerouting algorithms analyze background content to automatically adjust vector paths, reducing manual editing effort and improving accuracy.

DuoSpaceNet merges BEV and PV features to resolve the trade-off between computational complexity and 3D detection precision.

A vehicle detection apparatus divides image search regions into upper and lower zones using distinct optical flow models to identify forward vehicles.

Transmitting micro-Doppler measurements enables receiving updated AI models that improve object recognition accuracy while reducing signaling overhead.

Modifying generic images with imaging parameters creates realistic training data, reducing manual labeling time.

A hyperspectral image dimension reduction method establishes basis vectors to decompose spectral data into reduced dimension vectors for each pixel.

A system calculates vegetation index values from time series imagery to generate trend lines for automated crop monitoring.

A narrowband optical filter isolates specific light wavelengths to detect colored elements within an enhanced vision system.

Front camera free space signals identify close cut-in vehicles, reducing collision risks by improving recognition time.

A neural network rigidifies important synapses to maintain previously learned tasks while learning new ones.

A vehicle guidance system processes camera images and steering angle data to generate dynamic trajectory overlays for driver assistance.

A digital measurement system automates carbon stock estimation using satellite imagery and official databases.

A deep learning algorithm classifies flooded urban and rural areas using SAR imagery and an urban mask.

Local anomaly detection filters sensor data before sending intervals to a server, reducing manual review time while maintaining classification reliability.

Processor merges local camera images with remote vehicle sub-images to generate an expanded around-view display.

A traffic light prediction method associates lane line and obstacle data to generate topology information for autonomous vehicles.

Feature identification device detects separation poles and wire ropes using vehicle-mounted camera images for autonomous driving.

Server recognizes target objects and determines driving pattern priorities to enhance vehicle control stability.

Depth data analysis overlays obstacle distance on video feeds to resolve blind spot visibility limits.

Hybrid scale voxelization balances inference time and detection accuracy by segmenting point clouds into multiple scale groups.

Crossing transfer unit extracts regional and long-range spatial features from urban remote sensing sub-images for accurate scene classification.

A server device acquires captured images of a vehicle's exterior to identify unsafe driving events using a learned model.

Early sensor fusion defines regions of interest in camera images, reducing processing time while improving detection accuracy.

Dividing images into cells classifies ground and target features, resolving incomplete detection of objects near the ground while reducing computational effort.

A neural network generates modified bounding boxes to standardize annotations across diverse training datasets.

Dynamic classifier selection based on real-time rain intensity improves object detection accuracy where standard deraining algorithms fail.

Unsupervised clustering of satellite imagery identifies representative crop pixels to guide targeted ground truth data collection.

A key terrain finder system applies general, override, and user-defined scoring rules to generate an aggregate mask for terrain identification.

Autonomous vehicle system assigns interaction labels to nearby objects and generates cost functions for trajectory planning.

Modulated LED signals replace radio positioning, resolving location errors in crowded environments.

Starburst algorithm projects rays from point cloud clusters to generate shapes, reducing LiDAR processing power needs.

A vehicle monitoring system detects safety concerns while deactivated and notifies the driver upon reactivation.

A deep neural network system curates images using visual templates and self-supervised learning to generate refined image patches.