Abstract: In the traditional belt conveyor coal flow detection device, the nuclear belt scale has certain safety and environmental protection hidden dangers, and the detection precision of electronic belt scale is easily affected by the factors such as belt tension and stiffness. Moreover, non-contact detection methods based on technologies such as ultrasound, linear laser stripes and binocular vision have problems such as poor real-time performance and large measurement errors. A coal flow detection method for conveyor belt based on time-of-flight(TOF) depth image restoration is proposed. The TOF camera is used to obtain the coal conveying image of the conveyor belt. The TOF image is equalized, and the frame difference method and the boundary following algorithm are used to remove the background noise and obtain the coal region of interest. In order to solve the problem of inaccurate edge information caused by flying pixel noise and multi-path error noise at the edge of TOF depth image, the intensity image-guided depth image restoration algorithm is proposed. The Canny edge detection algorithm is used to find similar edges between the depth image and the intensity image. Based on the effective edge information of the intensity image, the unreliable data of the edge of the depth image is corrected. Furthermore, the high-precision depth images are obtained based on Navier-Stokes equation and median filter. The coal area is divided at the pixel level, the coal volume calculation model is established to obtain coal flow of conveyor belt by combining the conveyor belt speed. The experimental results show that the detection error is less than 3.78%, the standard deviation is less than 0.491 and the average processing time is 83 ms, which meets the actual production requirements.