High resolution depth sensor close-up imaging based on artificial intelligence has set a new record

     Kyocera Corporation has unveiled an AI-based, high-resolution depth sensor that utilizes traditional depth sensing technology to measure tiny objects that are difficult to measure. Its new camera provides a record depth measurement with a resolution of 100μm at a range of 10 cm, even for reflective or translucent objects. The company's innovations will support various fields such as manufacturing, medicine, logistics, etc. that require automatic recognition and precise depth measurement, unlocking the potential of artificial intelligence and robotics with visual capabilities far beyond those of the human eye.

     The unique configuration of the company's AI-based depth sensor includes two lenses on a single sensor, providing the industry's highest resolution depth measurement in stereo cameras to date. The extremely narrow baseline of its depth sensor allows it to calculate the difference in the position of objects through the left and right lenses at a shorter distance than traditional methods. This precision allows the sensor to make precise measurements of even the smallest objects.

     Artificial intelligence stereo vision algorithms allow accurate depth measurements of reflective or translucent objects. Traditional stereo vision algorithms involve object matching between left and right images. However, reflective or translucent objects often lack the desired contrast or are difficult to identify as the same object, resulting in measurement errors. The company's AI-based approach utilizes vast amounts of training data to accurately measure even challenging reflective or translucent objects.

     Although AI-based methods have many advantages, they usually require high labeling costs and long training times due to the large amount of data required to achieve high accuracy. As a result, the company has developed two key technologies to reduce training costs: label-free pre-training and the use of computational graphics to generate data, which has a tenfold increase in measurement accuracy.

     The company's unique AI solution uses label-free pre-training technology to provide equivalent recognition with only 10% of training data. In order to solve the problem that traditional AI requires a large amount of training data, it has developed a computer graphics data generation technology. This technology allows the automatic generation of training data in a CG simulation environment to accurately reproduce target objects and Settings. Improved and accelerated computational methods for CG rendering. This CG simulation for AI training allows it to adapt to new objects and environments, even reflective or translucent ones, for high-precision 3D distance measurements.