摘要： 针对水产养殖实践中鱼类摆尾频率计数困难的问题，提出一种基于计算机视觉的游泳型鱼类摆尾频率测量方法。利用摄像头获取视频，通过背景减法、二值化处理后得到只包含鱼的二值图像，并对二值图像细化得到鱼体脊椎方向的中线，再使用角点检测算法提取头部特征点、尾部特征点和鱼体脊椎曲线上特征点，进而通过特征点计算鱼体曲率。人工选取摆尾曲率无限接近零的图像，使用算法计算曲率数值，统计误差的大小和方差，用以确定统计摆尾次数的曲率阈值；方法的验证以大黄鱼（Larimichthys crocea）为实验对象，比较算法测量与人工计数的结果发现，正确率达到91.7%，能较好的测量摆尾次数；与基于距离的摆尾测量方法相比，该算法在测量曲率较大的图像时波动较小，更适合摆尾频率测量。

关键词: 游泳型鱼类, 计算机视觉, 摆尾频率

Abstract: In this paper, aiming at the difficulty in counting the frequency of fish tail beats in aquaculture practice, a computer vision-based method for measuring the frequency of swimming fish tail beats is proposed. The camera is used to obtain video, the binary image containing only fish is obtained after background subtraction and binarization and then thinned to obtain the median line in the direction of the spine of the fish, the corner detection algorithm is used to extract the head feature points, tail feature points and the feature points on the spine curve of the fish body, and the curvature of the fish body is calculated through the feature points. The image with the curvature close to zero is manually selected, and the curvature value is calculated using the algorithm. The magnitude and variance of the statistical error are counted to determine the curvature threshold. The method is verified by using Larimichthys crocea as the experimental object. Comparing the results of algorithm measurement and manual counting, it is found that the accuracy reaches 91.7%, which can better measure the number of tail beats. Compared with the method based on distance, the algorithm has less fluctuation when measuring the image with larger curvature, and is more suitable for the measurement of frequency of tail beats.

Key words: swimming fish, computer vision, frequency of tail beats