Scene Recognition TRIZ Cases
- Obstacle Cost Map Expansion for Safer Route Planning
Graduated obstacle costs expand binary maps into safer route fields, improving collision avoidance and route stability for mobile platforms.
- Neural Network Group Switching for Multi-Environment Accuracy
Setting-specific neural networks are selected and cross-checked to avoid accuracy loss and catastrophic forgetting in autonomous machine control.
- Fused Sensor Attribute Detection for Autonomous Vehicle Object Tracking
Bayesian fusion of camera, LiDAR, and radar probabilities improves object attribute detection for safer autonomous vehicle control.
- Vehicle Proceedable Direction Detection From Complex Traffic Signals
A camera classifier estimates proceedable certainty by bearing from traffic lights, arrows, signs, and road context to guide vehicle direction.
- Smartphone Remote Parking Using Camera-Based Distance Estimation
Camera-based vehicle image sizing lets a smartphone permit remote parking only within a legal operating range for safer external control.
- Mobile Robot Security Patrol with RFID Scanning and Semantic Mapping
A patrol robot combines RFID scanning, infrastructure communication, and semantic map updates to detect security violations and adapt safely indoors.
- Unmanned Transport Vehicle Control Using a Pseudo-Reference Position
A pseudo-reference position lets an unmanned transport vehicle move combined cargo more efficiently while avoiding obstacle contact.
- Neural Network Prediction Screening for Adversarial Perturbations
Gradient-based measurement quantities flag adversarially perturbed sensor inputs so neural network predictions can be rejected or warned on.
- Predictive Drone Pre-Surveillance for Targeted Property Monitoring
Behavior-based rules dispatch drones along predicted paths to pre-surveil property areas, improving safety without continuous monitoring.
- Pedestrian Activity Recognition With Meta-Learning and Limited Data
Using Siamese meta-learning and unsupervised clustering, this case cuts annotation effort while improving pedestrian intent prediction for vehicle maneuvers.
- Radar Neural Network Landing Guidance for Low-Visibility Runways
Fleet-shared radar image learning improves runway positioning and landing guidance when low visibility limits ILS, GPS, and infrared aids.
- Autonomous Vehicle Training Data Labeling Using Removable Sensor Pods
Removable sensor pods on additional vehicles automate synchronized labeling of autonomous driving data, cutting manual effort and label errors.
- Vehicle Environment Mapping with Radar Concealment Filtering
Radar mapping excludes concealed objects, removes polygon spikes, and adds virtual objects to cut false detections around vehicles.
- Video Display Using LoS-Matched Archived Segments in Low Visibility
Previously captured video with line-of-sight metadata is matched to a current region of interest to improve visibility in poor weather or light.
- Vehicle Surroundings Detection Using Radar Distance-Velocity Analysis
Radar distance and Doppler velocity data are combined into D2 distributions to detect stationary roadside objects without angle sensitivity.
- Steering Support Threshold Control for Multi-Object Avoidance
Dynamic steering-assist thresholds account for objects on both sides, avoiding torque that conflicts with driver intent during lane keeping.
- Autonomous Vehicle AI Training With GAN-Based Synthetic Scenes
GAN-generated objects are inserted into real scenes and artifact-cleaned to create accurate training data for autonomous vehicle detection.
- Optical Obstacle Sensor Subsystem for Cliff and Wall Detection
Projected indicator light and image analysis let self-guiding machines measure obstacle distance early enough to avoid wall collisions and cliff falls.
- Aerial Bin Imaging for Faster Inventory Content Verification
An automated aerial vehicle images warehouse bins and uses confidence-based vision analysis to cut manual inventory checks and errors.
- Data Center Task Scheduling Using Predicted Renewable Power
Camera-based sky analysis predicts solar power output so data centers can schedule workloads around renewable energy variability and cost.
- Point Cloud Free Space Estimation With Distance-Based Grid Probabilities
Distance-based point cloud zoning and cell probabilities improve autonomous vehicle free space estimation under sensor noise and availability limits.
- Vehicle Sensor Data Labeling via 2D-3D Track Aggregation
Automatic 2D-to-3D track aggregation labels repeated vehicle sensor captures faster, reducing manual annotation for object models and HD maps.