Scene Recognition TRIZ Cases

Driver history and situational context are used to suppress redundant vehicle alerts and set warning intensity for identified risks.

Deviation-based risk regions and adaptive prediction cycles improve future moving-body risk estimation while reducing vehicle processing load.

Camera images from the passenger-side mirror are shown when turn signals are active, preserving the driver's outside view despite cabin obstructions.

Real road images combined with simulated debris expand AV training data, improving object detection and hazard-aware navigation.

By detecting exit status and driving direction, the controller selects the relevant speed limit sign and avoids false ISLA warnings at exits.

A GAN discriminator flags synthetic top-down driving scenes, enabling broader autonomous vehicle training without costly real-world data capture.

Camera and lidar inputs feed a generic detector and selective classifiers to recognize regional road signs with lower processing load.

Uncertainty-aware object detection and prediction helps autonomous vehicles avoid hard braking and unstable motion plans when sensor outputs are ambiguous.

Reflected wave intensity lets vehicle radar separate lanes from road surfaces in rain or darkness, improving lane detection reliability.

A joined motion model predicts tractor-trailer states together, cutting compute load while improving autonomous vehicle trajectory accuracy.

Driving-task-based sensor classification prioritizes critical vehicle sensors to maintain reliable real-time detection with lower processing load.

Driver gaze and vehicle detection are combined to gate autonomous lane changes with stricter checks when another vehicle is in the target lane.

Neural and rule-based sensor fusion adapts camera, radar, and lidar inputs to improve object location reliability in changing driving conditions.

Multiple camera views and whole-image detection identify pedestrians on raised walkways or behind loads, improving driver alerts.

By extracting only specified sign characters such as numerals, vehicles cut image data volume, speed sign identification, and reduce transmission load.

Predicted occupancy grid maps let autonomous driving planners propagate uncertainty and learn safer trajectories from human driving data.

Size-based anchor sets split four-wheel vehicles, pedestrians, and two-wheelers to improve driving object detection accuracy with lower compute.

Brake light status changes the stopping distance to balance collision avoidance with smoother vehicle start and acceleration.

Environmental sensors and map data detect traffic signals ahead, letting longitudinal control adapt to driver attentiveness at intersections.

A feedback loop verifies only high-collision-risk objects, improving intervention reliability while controlling vision processing time.

Past front, rear, and side camera images are combined to generate underfloor views during turns while limiting image-processing load.

When camera lane detection degrades in weather or poor road scenes, stored images and neural prediction keep steering control stable.

Two-stage ROI prediction refines AGV trajectories with map and live environmental data to cut processing time and onboard compute demand.

Map-based confidence checks and sensor feedback help autonomous vehicles verify traffic light state and location despite obstructions and false positives.

An RSSAC control approach uses entropic risk and trajectory prediction to avoid collisions in dynamic environments with real-time efficiency.

Machine-learned error distributions for object position, size, pose, and velocity enable safer vehicle buffers and more realistic simulations.

A convex polygon built from predicted front-bumper travel improves collision detection accuracy while limiting buffer-region processing load.

When merge lanes are short or obstructed, lane-boundary following helps vehicles gain more time and distance for safer, smoother merging.

Sensor feedback and wiper position control prevent blade occlusion, preserving object detection accuracy for autonomous driving.

Timing-based AEB suppression uses front-vehicle entry estimates to avoid unnecessary braking when crossing traffic is detected at intersections.

Continuous existence-probability vectors fuse redundant sensor evidence to cut false braking and missed hazards in automated driving.

Machine-learning risk evaluation estimates unrealized collision probability by modeling how other road users may change trajectory.

Cameras and proximity sensors map vehicle flanks with a virtual distance grid to warn of boundary intrusion, contact risk, and door hazards.

Predicted object motion is converted into safe time windows for target positions, reducing replanning load while improving collision avoidance.

Stored baseline sensor data enables real-time event detection and alerts without streaming all raw data, reducing processing load and energy use.

Candidate scoring and staged classification help autonomous vehicles detect traffic light states accurately while limiting image-processing load.

Thermal cameras and temperature sensing detect a person in the garbage truck hopper and automatically stop compaction to prevent injury.

Image-based control activates vehicle object sensors only when free space and relative motion indicate need, reducing ultrasonic interference.

Variance-scaled initialization for depth-wise convolution keeps gradients stable when input and output counts differ, speeding convergence.

Correlation imaging with binary patterning and random grayscale data boosts frame rate while preserving image quality in vehicle lamp sensing.

A two-stage parking route uses a temporary target and mid-route switching to improve narrow-space parking accuracy and steering use.

Color and shape filtering narrows traffic sign search regions, cutting real-time computation while improving detection of damaged or dirty signs.

Staggered transmission of recognition images from multiple vehicle cameras cuts peak bandwidth and improves reliable reception.

Constrained EKF bounds on centripetal acceleration improve target curvature estimates, path prediction, and collision avoidance in ADAS.

Tracks a front vehicle's motion against its expected path to detect potholes, bumps, and diversions without costly high-resolution sensors.

Offset-aware hidden states align moving radar features across time steps, improving object detection and grid segmentation at high speed.

Synthetic fog images train a model to detect meteorological optical range quickly, so vehicles can ignore unreliable sensor data beyond visibility.

Front and rear lane curvature plus obstacle and vehicle-state data improve lane change decisions and automated steering or speed control.

Trajectory, road-segment, and height checks screen out irrelevant objects so vehicle planners reduce onboard computation without missing path conflicts.