Artificial intelligence 3D sensing image processing system based on array sensing Lidar aims to construct 3D image with high-quality immersion for AR/VR. The developed 3D scene recording system is based on the color camera and high-accuracy chaotic LiDAR. The chaotic Lidar with APD array and TOF sensors supports millimeter-accuracy and interference-avoiding capability within 100 meter in both indoor and outdoor environments. High-performance embedded CNN processor supports high-throughput, high energy-efficient, and low DRAM bandwidth computations for various image AI applications.

The 3D image recording system adopts Chaotic LiDAR to capture high-accuracy depth image with 1550nm eye-safe APD array and ToF circuits. By including RGB camera, new calibration/alignment and AI-based super-resolution techniques can increase spatial resolution of depth image and construct 3D image. Compared to the advanced Intel RealSense L515, the proposed system achieves higher accuracy (<1cm＠20m vs 1.55cm＠9m) and angular resolution (0.0125°x0.0125° vs 0.073°x0.071°) in both indoor/outdoor environments with an efficient eCNN core (7TOPS/W) for various image/video AI processing applications.

The proposed 3D image recording system can be applied to AR/VR real scene recording, health and sport technology, security, cyber physical system, etc. The Chaotic LiDAR can be also applied to vehicular sensing and 3D scanning products. Besides the APD array devices can be applied to optical-to-electrical converters in optical communications and various LiDAR systems. The proposed edge-computing eCNN processor can be applied to TV set with AI image processing acceleration functions such as super-resolution, image denoising, and style transformation, operating at 4K UHD 30fps specification.

線上展網址：
https://tievirtual.twtm.com.tw/iframe/ba4f360f-cb26-4646-8fe6-7b82aaa2619f?group=23bfb1fa-dd5b-4836-81a1-4a1809b1bae5&lang=en