Abstract: In the flotation process，the traditional monocular two-dimensional visual detection ignores the bubble depth motion component，which has the problems of velocity measurement distortion，poor adapt⁃ ability of complex pulp conditions，and low segmentation accuracy of adhesive bubbles. In order to realize the quantitative detection of full-dimensional parameters of primary bubbles in pulp phase，a three-dimen⁃ sional characteristic detector of pulp phase bubbles in flotation machine based on binocular vision was devel⁃ oped. By customizing the large depth of field binocular high-speed imaging hardware system adapted to the harsh working conditions of the slurry，the joint image preprocessing process of NLM non-local mean de⁃ noising + CLAHE contrast enhancement was constructed. The YOLO11m-seg lightweight instance segmen⁃ tation model was used to realize the pixel-level recognition of dense adhesion bubbles. Combining BoT�SORT multi-target cross-frame tracking and mask-constrained SIFT binocular stereo matching algorithm，a three-dimensional positioning and velocity vector solving model of bubbles was established，and the regula⁃tion law of working conditions was analyzed in combination with flotation multiphase flow hydrodynamics. The results show that the bubble segmentation accuracy of the fusion algorithm is 95.7%，and the false de⁃ tection rate is only 2.9%，which is much higher than that of the traditional watershed and Canny algorithm. The average relative error of binocular three-dimensional velocity measurement is 12.7%，which is much higher than that of monocular optical flow method. The industrial test shows that the impeller speed is the dominant control parameter of bubble motion，and the aeration amount only plays an auxiliary fine-tuning role. The effective bubble diameter of flotation is concentrated in 3~8 mm，and the roundness of large bub⁃ bles with a diameter of 12 mm is reduced to 0.67. The bubble size and shape are strongly related to the sepa⁃ ration index. The detector can completely output full-dimensional dynamic and static parameters such as three-dimensional velocity vector，particle size distribution and roundness of bubbles，and provide reliable three-dimensional data support for flotation kinetic mechanism research，fine adjustment of process parame⁃ ters and development of flotation intelligent control system.

Key words: binocular vision, flotation machine, bubble characteristic detection, three-dimensional ve? locity measurement, YOLO11 instance segmentation, BoT-SORT motion tracking