A mobile real-time detection method for 3D crabs oriented to 2D images

Abstract

The invention discloses a mobile 2D image-oriented 3D river crab real-time detection method. The method uses a CMOS camera installed under an automatic baiting boat to take a bird's-eye view, moves with the boat, and continuously collects underwater images that can reflect the natural living state of the river crab. Two-dimensional RGB images, and use annotation tools and augmented reality session data to jointly establish a dataset of river crabs with labeled poses and variable shapes; inspired by the anchor-free mechanism, using Gaussian center distribution only starting from a single RGB image, in the encoder‑ After the decoder architecture extracts features, it adopts a multi-task joint learning method supplemented by shape, detection and regression to predict the 2D bounding box of the unknown crab; and then through the perfect pose estimation algorithm EPnP, the 2D prediction is extended to the 3D bounding box to estimate the crab's Attitude and physical size, so as to build an ultra-lightweight single-stage 3D crab detection model. The method can improve the variable feeding efficiency and feeding effect of the automatic feeding boat.

Description

technical field [0001] The invention relates to the field of machine vision and pattern recognition, in particular to a three-dimensional detection and size measurement method based on two-dimensional machine vision. [0002] technical background [0003] River crab is the largest freshwater crab in my country, with rich nutritional and medicinal value, bringing huge economic benefits to fishermen. However, the current river crab farming is faced with the big problem of variable and uniform feeding, and its bait utilization rate and feeding efficiency are relatively low. This is mainly because the farming method of river crabs mainly relies on the fishermen's breeding experience to roughly estimate the amount of bait feeding, and does not Consider the huge differences in bait required across ponds due to differences in pond crab size, density distribution, and water environment. However, it is difficult to estimate the size and density distribution of pond crabs without huma...

AI Technical Summary

Problems solved by technology

Although there are many potential advantages in fish / crab classification and marine life monitoring, such methods are disadvantaged due to their high price, computational complexity and slow operation speed (compared to 2D methods) and the support of additional hardware. Not yet popularized in mobile devices (such as automated bait-casting boats)

Images