Automated detection of foot contacts, aerial phases, and visibility of bird

These codes were written by Daniel Hasse and were used, in combination with other methods, to perform an automated large-scale study (over one million frames) on the fluctuations of the CoM’s mechanical energy in several bird species (quail, oystercatcher, northern lapwing, pigeon, and avocet) during treadmill locomotion. see Andrada, E. et al. Mixed gaits in small avian terrestrial locomotion. Sci. Rep. 5, 13636; doi: 10.1038/srep13636 (2015). for more information events.R includes: Touch-down detection: we found that when the pixel values of the leg of every frame are projected onto the x-axis, the obtained 1D-curve display at TD a maximal bimodality. We measured the bimodality value of the 1D-curve by comparing it to a normal distribution of same mean and variance using histogram intersection distance. Frames in which a TD event occurs are then found by identifying all local maxima of the bimodality score over all frames (Andrada et. al., Mixed gaits in avian locomotion, in review, Fig. 5). Aerial phase: for each frame of a trial, we computed the minimum distance between the lowest point of the legs and the bottom of the image. Afterwards, a threshold was used to classify each frame as "aerial phase" or "stance phase" based on its leg height. For each sequence, the threshold was estimated automatically in such a way that only frames in which the legs are substantially higher than the median leg height are classified as "aerial phase" (Andrada et. al., Mixed gaits in avian locomotion, in review. Supplementary Information, Fig. S1). visibility.R includes: Visibility: to account for cases in which the birds left the field of view, for every frame of each trial we computed the number of pixels in which the bird is visible ('bird-pixels'). % of congruity was computed, only when the measured number of bird-pixel during all frames of a stance phase was higher than 80% of the maximal computed number of bird-pixels of the complete trial.

本代码由Daniel Hasse编写，可与其他方法结合使用，用于开展自动化大规模研究（涵盖超100万帧图像），分析多种鸟类（鹌鹑、蛎鹬、北凤头麦鸡、家鸽及长脚鹬）在跑台运动（treadmill locomotion）过程中其质心（Center of Mass, CoM）的机械能波动情况。更多信息可参考Andrada E.等人的研究：*Mixed gaits in small avian terrestrial locomotion*，发表于*Sci. Rep.* 5, 13636；DOI: 10.1038/srep13636 (2015)。

events.R 模块包含以下内容：
1. **触地检测（Touch-down Detection）**：研究发现，将每帧图像中腿部的像素值投影至x轴后，所得一维曲线（1D-curve）在触地（Touch-down, TD）时刻呈现最大双峰性（bimodality）。通过将该一维曲线与具有相同均值和方差的正态分布进行对比，利用直方图交集距离（histogram intersection distance）计算其双峰性数值。随后，通过识别所有帧中双峰性得分的局部极大值，即可确定发生触地事件的帧图像（Andrada et al., *Mixed gaits in avian locomotion*, 待刊, 图5）。
2. **空中阶段（Aerial Phase）**：针对每次试验的每一帧图像，计算腿部最低点与图像底部之间的最小距离。随后基于腿部高度设置阈值，将每帧图像划分为“空中阶段”或“支撑阶段（stance phase）”。对于每个图像序列，阈值将自动估算，仅将腿部高度显著高于腿部高度中位数的帧归类为“空中阶段”（Andrada et al., *Mixed gaits in avian locomotion*, 待刊, 补充材料, 图S1）。

visibility.R 模块包含以下内容：
**可见性分析**：为处理鸟类离开视野的情况，我们针对每次试验的每一帧图像，计算可见鸟类区域的像素数（即“鸟像素（bird-pixels）”）。仅当某一支撑阶段所有帧的鸟像素数高于完整试验中最大鸟像素数的80%时，才会计算一致性百分比（% of congruity）。