Abstract: To measure the 3D phenotypic parameters of Hyriopsis cumingii under dynamic transportation, we proposed a method based on oriented bounding box, keypoint detection and adaptive multi-frame robust fusion. Built on YOLO v8n-OBB and YOLO v8n-Pose, the method constructs 3D spatial mapping via color-depth registration and pinhole imaging model, and tracks dynamic individuals with DeepSORT. Incorporating edge integrity judgment, 5-frame median filtering and global sliding window optimal selection, it effectively suppresses angle jump and geometric misconnection. Experiments show that the method stably output phenotypic parameters of H. cumingii under multi-object dynamic conditions. Compared with single-frame averaging, the optimal 15-frame strategy reduced the MAE of shell length, full height, shell height, radial rib length and shell width by 13.0%, 45.8%, 30.0%, 13.0% and 16.8%, respectively, in angle-jump-prone postures; under random placement, MAE of all parameters decreased by 10%–30%. The average error of 5 parameters was about 2 mm, with a processing frame rate of 18.94 fps, satisfying real-time dynamic measurement needs. This method realizes 3D dynamic phenotypic measurement via visual-depth fusion and temporal adaptive measurement, providing a high-precision, portable tool for aquatic shellfish breeding and phenotyping, and technical support for non-contact measurement, individual identification and phenotypic database construction of aquatic animals.