一种简单有效的鱼群轨迹追踪算法

A fast and effective algorithm to track fish school

在鱼群运动机制的研究中,由于鱼群具有不同的体征、复杂的运动模式和运动中频繁的遮挡,所以如何获取鱼群个体的时空轨迹是一个非常重要而且困难的问题.为了获取鱼群的轨迹,本文提出了一个简单有效的方法来侦测和跟踪鱼群.整个追踪算法包括视频输入、图像获取及预处理、目标检测、数据关联、数据输出和人工校正6大模块.算法通过和Id tracker,Ctrax对于不同实验环境视频的轨迹提取结果的对比,表明了本文算法在准确度、识别效率及适用性上的优越性.

To investigate collective motion of groups of animals, it is important to track multiple individual moving animals and acquire their positions over time and space. A few studies have tried to solve this problem aiming for automated data acquisition. But none have solved the problem adequately, since automated tracking is difficult to achieve due to complexities in individual shape, sophisticated motion patterns and frequent occlusion. Several algorithms on this problem have been published, usually for one special species, the zebra fish, for instance. Such algorithms tended to be very demanding regarding the video quality （high frame rates, high image resolution and steady background）, and often are very time-consuming. Here we have developed an integrated approach based on artificial neural networks which enables us to automatically extract individual trajectories from both high and low quality videos. We applied our method to track different fish videos, it was found that our method has a high efficiency and accuracy in most situations.