Abstract: Abstract: To address the problem of non-contact volume measurement of coal gangue in dynamic conveyor belt scenarios, where the irregular shape and high-speed movement of coal gangue pose challenges, this paper proposes a coal gangue volume estimation scheme based on an improved projection integration method from the perspective of computer vision and point cloud processing. The method takes a monocular line laser camera as the core of data acquisition, completes 3D point cloud reconstruction through high-frequency profile scanning combined with conveyor belt movement information, and constructs a full-process computer vision processing chain of "preprocessing - segmentation - completion - integration": the RANSAC algorithm is used to robustly fit the conveyor belt plane to achieve background elimination; the region growing algorithm based on normal vector and curvature constraints is used to complete the instance segmentation of coal gangue point clouds; the normal foot projection and concave hull/elliptical contour fusion strategy is designed to realize the completion of the bottom structure; finally, by introducing concave hull boundary constraints, adaptive grid division and radial sector parallel integration mechanism, the volume calculation accuracy of the traditional projection integration method is optimized. Experimental verification shows that on 170 sets of coal gangue point cloud data, the average volume estimation error of this method is 8.92%, and 95.88% of the sample errors are controlled within 20%, which effectively im-proves the accuracy of non-contact measurement of coal gangue on the conveyor belt. This research provides a computer vision solution with both robustness and practicability for 3D volume measurement of irregular objects in industrial scenarios.