Scene Recognition TRIZ Cases

A monocular-camera world model uses environment memory and a nematode-like neural circuit to avoid manual modular redesign in autonomous driving.

Multi-sensor fusion combines GPS, cameras, magnetic markers, and proximity sensing to guide autonomous vehicles into parking spaces precisely.

Physics-based perception fields turn surrounding objects into virtual forces, reducing jerky ADAS behavior while improving explainability and robustness.

Separating the pixel array and neural-network DSP across bonded substrates cuts noise during exposure and readout while preserving image quality.

A two-stage neural network scores object centers in one pass, then batches property estimation to cut vehicle perception latency.

Occupancy grids and a single-stage neural network detect nearby vehicles in 10 ms while generating oriented bounding boxes for precise localization.

Uses turn-signal status and predicted oncoming paths to warn before a turn starts, reducing collision risk without excess alerts.

Filtered traffic scenarios train and validate models that infer pedestrian intent, improving autonomous vehicle motion planning and safety.

By learning what belongs on the road surface, this model flags unknown obstacles without manual class annotation or new data collection.

A road-aware gaze zone shifts and resizes with curves and lane changes to cut false driver-monitoring alerts while preserving attention checks.

Fusing camera, radar, and lidar data adds height and drivable semantics to grid cells, improving vehicle localization and obstacle handling.

Contour alignment around the optical center improves vehicle object tracking and time-to-collision estimation without LiDAR or radar.

When visible cameras fail in glare or darkness, infrared activation and image fusion improve ADAS object detection while limiting energy use.

Alternating visible-light patterns let a vehicle camera capture 3D shape and motion at night without the cost and complexity of multiple sensors.

Event-based sensing captures subtle object motion changes around vehicles to predict imminent movement faster and cut data bandwidth.

Alternating full- and subset-emitter LiDAR scans separate return events in time to remove crosstalk and reduce false detections.

During sidewalk pull-over, the vehicle stops only for obstacles in its travel path, avoiding unnecessary stops from side or rear detections.

Multiple sensor arrays check ground and overhead clearance so tall vehicles can identify suitable parking spaces and avoid height-related collisions.

Fusing camera and radar data, the ECU detects intersecting-lane traffic, calculates TTC, and brakes early to avoid cross-path collisions.

Maps high-dimensional vehicle sensor points into an abstract space to detect surrounding features with lower computational load and preserved context.

Visualized trajectory points, sensor frames, and editable constraints help validate and refine SLAM maps when GPS signals are blocked.

When sunlight degrades distance sensors, the controller shifts from adaptive to constant-speed cruise to preserve safe operation.

Within circuit areas, ROI-based image processing prioritizes signal flags and obstacles to improve driving guidance without full-frame delay.

Direct pixel-based prediction of target vehicle line crossing avoids world-frame calibration errors and cuts ADAS response lag.

Comparing predicted and observed object behaviors lets autonomous vehicles rank critical prediction errors by overlap and correct the most relevant ones first.

During lane changes, the controller compares camera and map lane lines, prioritizing real-time recognition when mismatches could disrupt safe control.

Event-driven pixel sensing detects subtle object motion before full frames arrive, cutting response time and bandwidth for collision avoidance.

Projected road-user and target-trajectory graphics help drivers grasp automated lateral control behavior and when manual intervention is needed.

Map-defined road boundary regions filter sensor ghosts from buildings and trees, reducing false moving-object detection and unnecessary deceleration.

Selective ROI masks and simplified boundary polygons speed occupancy grid updates while preserving object and drivable-area accuracy.

Time-series image analysis separates LED flicker from true flashing in traffic signals, improving AV signal state classification.

Object data shared from other vehicles extends detection beyond onboard sensor limits, improving autonomous navigation around curves and in bad weather.

Ranks nearby traffic lights by relevance and cross-checks their status under occlusion or failure to support braking and acceleration decisions.

Removes reflective ghost regions from vehicle camera images, then re-detects objects with reliability checks for steadier autonomous driving.

Multiple rotated or shifted scene views are batched through one model to cut flickering predictions and avoid ensemble-level compute costs.

Camera-derived feature vectors update a shared database so vehicles classify traffic signs accurately without storing full map and sensor datasets.

Co-located camera and LiDAR sensors use overlapping views in one housing to classify nearby objects and reduce vehicle blind spots.

Sensor fusion estimates whether a school bus is loading, unloading, or inactive so an autonomous vehicle can slow or stop compliantly.

Multiple encoder RNNs fuse LIDAR, RADAR, and camera sequences over time to improve object classification without manual feature tuning.

Onboard sensors assess hazards near autonomous vehicle drop-off points, enabling safer rider exit or alternate locations.

Obstacle residence time estimation helps an autonomous vehicle decide whether to wait or re-plan, improving real-time avoidance in road and open-space driving.

Merged LIDAR, camera, and motion data turn recorded drives into reusable simulation scenarios with static objects, road context, and road-user trajectories.

Candidate ROI scoring focuses vehicle image analysis on likely objects, boosting frame rate and cutting CPU load for path planning.

Camera-based parking guidance detects spaces and vehicle motion to reduce blind-spot errors and help drivers maneuver safely into parking spots.

Terrain mapping and vehicle clearance data are used to set speed, gear, and path guidance that help prevent off-road damage or getting stuck.

Grid-based assignment of detected road signal signs improves junction signaling identification and supports more reliable automated longitudinal guidance.

Wide-field noncontact sensing detects obstacles in a vehicle powered door path early enough to stop, slow, or reverse before contact.

Multi-sensor fusion of camera, infrared, and radar data predicts animal crossings early enough to trigger driver alerts and safer vehicle response.

Multi-sensor top-view fusion converts camera, lidar, and radar data into 3D stixels to detect free space and obstacles for vehicle control.

Computer vision checks garage door status before and after in-vehicle remote commands, avoiding GPS-trigger errors and confirming execution.