基于学习的斑马鱼检测与跟踪  

LEARNING-BASED ZEBRAFISH DETECTION AND TRACKING

作　　者：朱佩儒 刘烨[2] 王硕鸿 刘俊[1] 陈雁秋[1] 

机构地区：[1]复旦大学计算机科学技术学院,上海200433 [2]南京邮电大学自动化学院,江苏南京210023

出　　处：《计算机应用与软件》2015年第9期227-230,250,共5页Computer Applications and Software

基　　金：国家自然科学基金项目(61175036)

摘　　要：群体行为吸引了各个领域众多科学家的兴趣,而斑马鱼作为一种模式生物,被广泛运用为研究群体行为的对象。近年来,由于照相机系统和跟踪算法的发展,使通过分析鱼的运动轨迹来研究群体行为成为了可能。但是,怎样从视频中准确鲁棒地恢复鱼的轨迹仍然是一件非常具有挑战性的问题。为了解决该问题,提出根据鱼的头部特点和成像特征,设计一个基于统计学习的鱼头检测器,从而减少身体形变对跟踪的影响;同时,通过给斑马鱼的运动建模,结合全局匹配算法,使跟踪算法对漏检、错检和短暂的遮挡有很强的容忍性。大量的实验表明所提出的鱼头检测和跟踪算法的准确性和鲁棒性。Collective behaviours have attracted enormous research attention from scientists of various research fields, and zebrafish, as a kind of model organism, have been commonly used as the object for collective behaviour research. Recent advances in camera systems and tracking algorithms have made it possible to study their collective behaviours through analysing their motion trajectories. But how to recover the trajectories of fish from the video with accuracy and robustness is still a challenging task. In order to salve this problem, we design a sta- tistic learning-based fish-head detector according to the head characteristic and imaging feature of fish, so that reduce the influence of body deformation on tracking; Meanwhile by modelling the motion of zebrafish and combining global matching algorithm, the tracking method has strong tolerance on missing detection, false detection and short time occlusion. Excessive experiments demonstrate the accuracy and robust- ness of the proposed fish-head detection and tracking algorithm.

关 键 词：卡尔曼滤波 最优线性指派问题 支持向量机 斑马鱼 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]