Scene Recognition TRIZ Cases
- Hierarchical ML Network for Rare Object Detection
A hierarchical machine-learning network architecture refines object classification using specialized sub-class models and region of interest inputs.
- Automatic Driving System Reliability Comparison
An automatic driving system compares travel plan reliability to resume control based on the first plan.
- Spatio-Temporal Land Cover Prediction Using Dual-Memory LSTM
Dual-memory LSTM framework extracts discriminative features from satellite imagery to predict land cover changes.
- Point-Based Prediction System for Vehicle Scene Analysis
A point-based prediction system processes vehicle environment data using a neural network to generate object movement forecasts.
- ML Network Training for Vehicle Sensor Dropout Robustness
Simulating sensor drop rates during machine learning network training enables accurate environmental prediction despite missing image frames.
- Vehicle Object Detection Using Laser Radar Prediction
A vehicle image processing device uses laser radar to detect objects outside the camera view and generates predicted area information for faster recognition.
- Ultrasonic Object Detection via Dual Transceiver Intersection
Dual ultrasonic transceivers calculate object position by intersecting wave loci, resolving measurement precision limits in complex reflection scenarios.
- Pedestrian Tracking via Stereo Disparity and Body Part Segmentation
Segmenting pedestrians into body parts using stereo disparity enables accurate tracking when occlusions or posture changes obscure the full figure.
- Camera Performance Threshold Detection via Gradient Image Analysis
Gradient image analysis establishes camera failure thresholds by evaluating object recognition correctness, reducing extensive road test requirements.
- Automated Vision System Validation via Synthetic Ground Truth
Synthetic ground truth enables automated comparison of test and validation object detection outputs, identifying discrepancies without manual analysis.
- Multispectral Imaging for Atmospheric Chemical Species Detection
Multispectral imagery extracts chemical species concentrations through spectral band comparison and albedo normalization.
- Unified CNN for Autonomous Vehicle Sensor Fusion
A unified convolutional neural network processes multi-modality sensor data through shared layers, reducing training time and computational power.
- Autonomous Vehicle Gradient Detection via Sensor Fusion
Processor fuses camera and lidar data to calculate road gradient for precise route generation.
- Ultrasonic and LIDAR Sensor Fusion for Reliable Obstacle Detection
A dual obstacle detection device combines ultrasonic and LIDAR sensors to maintain continuous coverage when one unit becomes impaired.
- Deep Neural Network Confidence via Standard Deviation
Multiple deep neural networks estimate trailer angles while calculating prediction standard deviation to resolve accuracy versus complexity trade-offs.
- Multispectral Surface Contact Detection for Maritime Assets
A multispectral imaging system detects surface contacts by analyzing pixel variance anomalies and object motion relative to water waves.
- Distributed Fiber Sensing Traffic Monitoring with Neural Denoising
A distributed fiber-optic sensing system uses deep neural networks to denoise waterfall traces for accurate traffic data extraction.
- Obstacle State Detection via Multi-Sensor Fusion
D-S evidence combination fuses multi-sensor data to resolve measurement precision versus device complexity contradictions in unmanned vehicle systems.
- Plausibility Check Module for Driver Assistance Object Recognition
A plausibility check module compares sensor data with external reference information to classify vehicle surroundings.
- DNN Confidence Valves via K-Means Clustering
K-means clustering adjusts deep neural network confidence scores using distance ratios to resolve untrustworthy high values from limited training data.
- Dynamic Occupancy Grid Fusion for Autonomous Vehicles
Fusing radar velocity data with camera occupancy grids resolves measurement precision trade-offs in autonomous driving.
- Adaptive Point Cloud Augmentation for ML Training
A confusion matrix policy drives adaptive point cloud augmentation to inject deficient object instances into training frames.
- Cross-Modal Target State Detection via Segmented Feature Extraction
Segmenting camera and radar feature extraction reduces upstream noise accumulation, improving velocity and orientation prediction accuracy.
- Message Passing Network Object Signature Tracking
Graph-based message passing fuses multi-camera embeddings to resolve occlusion and view variation bottlenecks in autonomous vehicle tracking.
- Vehicle Surroundings Monitoring via Pitch Angle and Gradient Detection
Surroundings monitoring apparatus switches display output based on vehicle pitch angle changes to present captured image data.
- Object Recognition Using LIDAR Lane Geometry
This LIDAR object recognition system calculates straight lines from object box reference points against host lane boundaries to resolve positioning errors on curved paths.