Scene Recognition TRIZ Cases

This case uses object shape features, loss functions, and particle swarm optimization to calibrate lidar and IMU.

Machine learning combines remote sensing, elevation, and soil surveys to predict soil properties and guide variable-rate inputs.

A vehicle recognition unit detects concave upper-surface sections to split merged point clusters and identify nearby objects separately.

The image processor detects objects in overlapping projections and adjusts the reference plane to prevent blur or image absence.

Replacing Softmax with ReLU normalization and downsampling lowers compute and memory demands for attention-based 3D detection.

Class-based image matching reduces false abnormalities caused by weather and time changes.

A Bayer sensor reads wavelength from spread laser spots, helping detect varied power levels while limiting sensor damage.

Cross-modal image and point-cloud checks identify abnormal sensor data before it can mislead the automotive perception engine.

This case combines different image recognition models and overlap analysis to improve detection accuracy across varied object sizes.

Deep learning links workers across camera feeds and classifies PPE to flag safety incidents across construction sites.

Short- and long-term neural fusion separates frame histories to improve 3D detection accuracy while reducing compute demands.

An image classifier selects lane-containing regions before regression, improving coordinate accuracy and training stability.

A vehicle-mounted camera identifies arrow and line-segment critical points to reduce redundancy and improve sign semantics.