Kernan W J, Mullenix P J, Hopper D L
Department of Physics and Veterinary Diagnostic Laboratory, Iowa State University, Ames 50011.
Pharmacol Biochem Behav. 1987 Jul;27(3):559-64. doi: 10.1016/0091-3057(87)90367-4.
Analysis of animal behavior has been an arduous task requiring a human observer to record and classify individual motor acts. A computer pattern recognition system is introduced which simplifies this task by minimizing the need for human intervention. This system uses two video cameras with horizontal and vertical views of the behavior of a control and an experimental rat as they explore a simple environment for 15 minutes. Their behavior is sampled at a rate of one frame/second. Data from the video cameras are then converted into a form acceptable to Micro Vax I and VAX 11/750 computers. Each video picture is reduced to a 256 by 256 array, and ultimately each 15 minute observation session generates 28,800 blocks of information at 512 bytes each. Using a mathematically complete set of moments to the fourth order and the associated scalar invariants, the computer is programmed to identify the five major body positions of the rat including standing, sitting, rearing, walking and lying down. The computer also is programmed to identify the behaviors of grooming, head turning, whole body turning, looking, smelling, sniffing and washing face. This computer pattern recognition system not only speeds up behavioral classification, it alleviates the much criticized subjectivity introduced by human observers.
动物行为分析一直是一项艰巨的任务，需要人类观察者记录并对个体运动行为进行分类。本文介绍了一种计算机模式识别系统，该系统通过尽量减少人为干预来简化这项任务。该系统使用两台摄像机，分别从水平和垂直角度拍摄一只对照大鼠和一只实验大鼠在一个简单环境中探索15分钟的行为。行为以每秒一帧的速率进行采样。然后，摄像机的数据被转换成适用于Micro Vax I和VAX 11/750计算机的形式。每个视频画面被缩减为一个256×256的数组，最终，每个15分钟的观察时段会生成28,800个信息块，每个信息块为512字节。利用一套数学上完整的四阶矩及相关的标量不变量，计算机被编程以识别大鼠的五个主要身体姿势，包括站立、坐着、直立、行走和躺下。计算机还被编程以识别梳理毛发、转头、全身转动、张望、嗅闻、吸气和洗脸等行为。这种计算机模式识别系统不仅加快了行为分类的速度，还减轻了人类观察者引入的备受诟病的主观性。
