The invention discloses a giraffe
behavior recognition method based on a three-flow convolutional network, and the method mainly comprises the steps: carrying out the target detection and posture
estimation of each frame of image in a video through employing a
convolutional neural network, and obtaining the bounding box and key feature point coordinates of each giraffe; then, a multi-target tracking
algorithm is combined with the skeleton attitude similarity to realize cross-frame tracking and ID maintenance of individuals, for each tracked giraffe individual, a whole-body video clip with a conventional
frame rate, a frame reduction clip and a local action clip with the mouth as the center are extracted, the three types of clips are respectively input into different global
feature modeling networks, and the whole-body video clip, the frame reduction clip and the local action clip are subjected to
image fusion; and the giraffe
behavior recognition module is used for extracting multi-scale behavior feature vectors, realizing adaptive
feature fusion through the gating attention fusion module, and finally outputting typical behaviors through a full connection layer and a
probability model, thereby realizing high-precision giraffe
behavior recognition.