Obtaining high-quality color images and 3D point cloud information through a 3D visual perception system is the key to applications such as mobile robots. Since the existing 3D visual perception system is difficult to meet the requirements of real-time

high precision and high robustness at the same time

a binocular high-speed 3D visual perception system integrated with ToF(Time of Flight) is designed and realizedin this paper. The system is mainly composed of camera board

FPGA board

GPU board and ToF board. High-speed image acquisition and preprocessing are realized through FPGA board

and Fast-SGM algorithm adapted to FPGA is proposed to obtain high-speed binocular stereo matching depth map. Collect the ToF depth map and fuse the binocular stereo matching depth map with the ToF depth map based on the confidence

so as to solve the problem that the binocular stereo matching fails in the case of weak light and weak texture and the ToF fails in the case of strong light and long distance

to further improve the measurement accuracy and robustness. The system accelerates data processing through the parallel acceleration characteristics of FPGA and GPU

and can output dual-channel RGB images and stereo matching depth images with a resolution of 1 024 pixels×1 024 pixels and a frame rate of 120 fps

and can output a resolution of 1 024 pixels× 1 024 pixels

fused depth images with a frame rate of 60 fps

which can meet the needs of real-time applications such as mobile robots for environmental depth estimation.