Abstract: Coal gangue exhibits complex morphology, rough surfaces, and significant size variations. In dynamic conveying scenarios, it is easily affected by factors such as reflection, occlusion, and motion asynchrony, which lead to breakage or displacement of laser line stripes, resulting in point cloud sampling loss and volume measurement errors. To address this problem, a point cloud volume measurement method for coal gangue based on an improved projection-based integration method was proposed. The Random Sample Consensus (RANSAC) algorithm was used to fit the main plane, and a spatial filtering criterion was applied to effectively remove the conveyor belt background and noise. Initial region growing segmentation was performed based on normal and curvature constraints, and a multi-factor clustering mechanism was introduced to eliminate over-segmentation interference, thereby achieving accurate instance segmentation of adhesive objects. Considering that the vertical scanning perspective of the line-laser camera and the irregular natural morphology of coal gangue caused severe self-occlusion in the bottom region, a bottom surface completion strategy integrating normal foot projection and uniform density filling was proposed, and a closed bottom contour was reconstructed using a two-dimensional concave hull or ellipse fitting. In the traditional projection-based integration process, concave hull boundaries were introduced to eliminate redundant empty grids. The median criterion was applied to remove height outliers, and a radial-sector parallel strategy was adopted to improve overall computational efficiency and noise robustness. The experimental results showed that the overall average relative error of coal gangue volume measurement was only 8.92%, and the qualification rate reached 95.89% under the maximum allowable error standard of 20%. In multi-orientation flipping tests of the same gangue, the average relative error of volume measurement was only 5.7%.