Scene Recognition TRIZ Cases

Camera image embeddings flag anomalous road conditions so autonomous vehicles can navigate accurately with less mapping data and processing load.

Multiple cameras estimate a preceding vehicle's angle and lateral speed, enabling predictive path biasing to handle cut-ins more safely.

Pre-acquired longitudinal and transverse gradient data improves route following by tuning speed, ADAS control, and sensor offset correction.

Multi-stage image scoring and classification improves traffic light detection under changing conditions for safer autonomous vehicle control.

By fusing 3D human keypoints with trajectory context, this case improves pedestrian crossing and path prediction in complex urban scenes.

Image-based pedestrian attribute estimation adjusts braking and warning timing to better protect children and elderly people near vehicles.

Stereo-camera analysis maps risk areas around scattered road objects using luminance dispersion and height distribution for safer vehicle travel.

Point cloud and camera data are fused in a neural network to predict open vehicle doors and support safer autonomous trajectory planning.

Predicts obstacle motion from adjacent-lane vehicle effects to avoid excessive emergency maneuvers and improve collision avoidance.

Predictive tracking of nearby traffic objects keeps autonomous parking active in traffic jams instead of ending early and forcing driver takeover.

By tracking adjacent vehicle braking, turning, and speed changes, this case enables earlier blind-zone hazard response and collision avoidance.

Sensors track nearby people and objects while projected ground guides and external displays support safer vehicle entry, exit, and passage.

Occupancy flags mark tested and occupied grid cells so parkable spaces can be found faster without redundant cell checks.

Hierarchical pixel labeling combines 2D sensor data, confidence, and risk to cut false construction zone labels for safer vehicle navigation.

Synchronizing a lead vehicle's speed decrease with brake light timing cuts false detections from reflections and improves braking recognition.

Stereo cameras and radar estimate collision time, impact position, and post-collision travel range fast enough to help avoid secondary crashes.

KGNN-based graph linking and embedded descriptors improve monocular semantic keypoint tracking under lighting and viewpoint changes.

Time-synced detections from nearby AVs supplement long-range and occluded object labels, reducing manual labeling and improving training data.

Image and acoustic sensing capture another vehicle’s light pattern, convert it to metadata, and enable compatible personalized effects.

Displays surround images only when intersection visibility is restricted, reducing driver distraction while preserving blind-area awareness.

A two-cycle LiDAR scan alternates full-channel and subset emissions to identify crosstalk and improve point cloud reliability.

Iterative contour rotation refines parking space orientation from partial sensor data, enabling collision-free parking in angled spaces.

LiDAR-based dynamic occupancy grids assign cell-level occupancy probabilities to estimate time-to-collision and guide vehicle avoidance maneuvers.

Sensor-driven map adaptation converts non-HD maps into HD-compatible data, reducing dual-stack complexity for autonomous driving.

A unified token network replaces separate vehicle models to fuse multimodal driving data, improving prediction accuracy while cutting training complexity.

Tactile seat belt vibration warns distracted drivers of autonomous driving risk while reducing sensor-processing complexity.

Stored parking features are refreshed using recognition and illuminance matching, keeping automatic parking accurate as weather and surroundings change.

Sensor and pose-based gesture prediction helps autonomous vehicles anticipate human intent and plan safer trajectories with faster decisions.

Moving object traces generate pseudo-labels from unlabeled LiDAR sequences, cutting manual labeling while supporting accurate 3D object detection.

Structured object-level tokens let an autonomous driving LLM improve scene reasoning and route planning in long-tail scenarios.

A two-stage obstacle check refines boundary contours near a route to cut false warnings and improve collision monitoring reliability.

Rear-vehicle V2X sensing extends drop-off assistance beyond host sensor blind spots, reducing added sensor complexity during parking.

Optical gesture recognition activates running boards and compaction functions automatically, reducing manual strain and time loss on collection routes.

A modified area aligned to obstacle contours improves safe-area invasion checks, cutting false warnings in vehicle collision monitoring.

Satellite images combined with weather data estimate wide-area road surface conditions, including freezing and fallen leaves, for safer travel planning.

Camera frames are segmented into BEV polygons and center polylines to create shareable vectorized maps with lower mapping cost and faster updates.

Camera-radar defensive driving predicts adjacent-lane and obstacle motion to trigger braking, steering, or speed changes before collisions develop.

Combining multiple LiDAR point clouds with camera bounding boxes improves object labeling accuracy while managing processing complexity.

Selective backup camera views switch between wide, cropped, and processed images to reduce fisheye distortion while preserving parking visibility.

Sensor fusion builds a unified 3D vehicle surround view, reducing interface switching and helping drivers spot critical objects.

Aggregated camera, radar, lidar, and ultrasonic data build a unified 3D vehicle view that reduces display switching and driver overload.

Temporal feedback in a multi-task vehicle perception model improves dynamic environment representation while sharing features across detection tasks.

Reference-based confidence scoring monitors object perception reliability and lets ADS functions adapt vehicle control when sensor output is uncertain.

Driving-task-based sensor prioritization cuts processing load while preserving timely vehicle environment detection in complex scenarios.

A mobile object uses a first sensor plus contact-based confirmation to find hard-to-detect nearby objects and guide smoother movement.

Vehicle orientation narrows image analysis to likely door regions, helping autonomous vehicles detect open doors with less data processing.

Fusing scene embeddings from sensor outputs with image embeddings improves vehicle object detection and classification when sparse data misses details.

Earlier frame feature maps enrich current sensor features to improve 3D bounding boxes, object characteristics, and trajectory prediction.

Dynamic fluctuation thresholds tied to driver abnormality recognition help distinguish risky neighboring vehicle behavior from wind or road effects.

Reference-data comparison assigns confidence to vehicle object perception output, helping automated driving functions respond safely to failures.