It’s crucial to obtain the dimensions of fish fry accurately and quickly in aquaculture. Traditional manual sampling and measurement are time-consuming and labor-intensive and cannot meet the demands of smart aquaculture development. A vision-based method for rapid measurement of fish length is proposed for grass carp fry with length distribution from 20 to 100 mm in this paper. It allows for quick and accurate length measurement of test fish fry without reference and in a non-contact manner. Firstly, an RGB-D camera is used to capture depth information and gray images of the target. Those images are processed to segment the target fish from the background. For the case of overlapping fish, concave regions and points are extracted to separate individual fish based on concave points. An improved thinning algorithm is then used to extract the fish skeleton, and key skeleton points are selected. Finally, by combining the image depth information, the three-dimensional coordinates of the skeleton points are transformed，allowing for the accurate measurement of the total length of the fish fry. Experimental result shows that the proposed method achieves an average absolute error of 1.57 mm and an average relative error of 4.39% in the length measurement of the test fish. It provides a non-contact measurement method that supports applications such as graded feeding and intelligent feeding in the aquaculture industry.
