基于视频的果蝇求偶行为识别和运动轨迹跟踪预测

针对人工检测果蝇求偶行为效率低、果蝇轨迹跟踪困难的问题, 提出一种基于视频的果蝇求偶行为识别和运动轨迹跟踪预测的方法。首先采用Ostu法实现图像分割，并使用外接椭圆锁定果蝇目标，根据果蝇位置关系和差值图像灰度均方差进行求偶行为误判筛除; 然后提取果蝇轮廓的形状特征(面积，周长，似圆度等)和7个不变矩特征，并依据特征信息实现果蝇求偶行为的识别; 最后采用卡尔曼滤波和特征匹配的轨迹跟踪方法实现运动果蝇的跟踪预测。实验结果表明，该方法能够准确识别果蝇的求偶行为和运动跟踪预测。

Drosophila Movement Trajectory and Courtship Behavior Reappearance Based on Video Processing

A method of drosophila movement trajectory and courtship behaviorreappearance based on video processingis proposed to increase the manual detection efficiency of drosophila courtship behavior and track the drosophila movement trajectory more easily. After the image was segmented with Ostu algorithm, the drosophilas were located in circumscribed ellipses. Then shape characters(area, perimeter and roundness) and 7 moment invariant characters were achieved, and fake courtship drosophilaswere deleted according to positional relation and gray mean variance of the difference image, anddrosophila courtship behavior was recognized with the shape characters and moment invariant characters. Finally Kalman filter and feature matching method were used to predict drosophila movement trace. The experiment result shows that the method presented in this paper could recognize courtship behavior of drosophila and stalk moving drosophila accurately.