Three Technologies Used in 3D Data Analysis
Classification
The technology that recognizes and classifies 3D objects. It can differentiate between types and assign qualitative rankings.
Part Segmentation
A technology that automatically distinguishes between partial sections of the target. Capable of extrapolating information, etc. on required parts.
Semantic Segmentation
Objects are extracted from an image, and the AI recognizes the boundary outline for each object. With aerial 3D landscape data, the AI can analyze information about the area.
case.01Optimizing performance using a
self-flying drone and AI
We developed the drone camera “Blast Eye” for viewing blast sites in real time, and developed “Blast AI” to study the observations and judgments of experts in order to determine the quality of a site.
When construction company Toda Corporation and Rist Co., Ltd. excavate mountain tunnels by blasting through the rock, the next blasting pattern is determined depending on the shape of the blasting stone (blasting stepping stone) from the last blast. Up until now, those judgments were made by experienced tunnel technicians, but we developed the “Blast Eye / AI” that automatically judges the quality of blasts by studying and learning from those experts’ decisions.
Blast Eye
Stable flight even in long, narrow stretches where GPS cannot be used.
Blast AI
Studies various simulated blasting stone shapes and learns to determine blast quality.
case.02Utilizing 3D laser radar AI
Real-time 3D information acquisition and object recognition, behavior analysis, and shape recognition technology using 3D-LIDAR
Makes a real time measurement of the amount of time it takes for the LIDAR laser to be reflected back from the target object, and obtains a 3D image of the object.
Example of motion detection using 3D laser radar
Simultaneously recognizes both the presence of pedestrians and the motion vector (detecting the direction in which it is moving and its speed)
https://www.konicaminolta.com/jp-ja/future/3dlr/
https://special.nikkeibp.co.jp/atcl/NBO/16/010500001/
Estimating the distance to low-reflectivity objects with deep learning
Improving distance measurement accuracy for low-reflectivity objects, which is a challenge for 3D-LIDAR
Black cars are recognized, distance information is supplemented, and the exact distance is calculated.