Animal group trajectory tracking method and system

Abstract

The invention relates to an animal group trajectory tracking method and an animal group trajectory tracking system. The method comprises the following steps: S1, feeding an animal group on an environment platform suitable for survival, activities and social contact of the animal group; s2, collecting a monitoring video of the social interaction behavior of the animal group through a camera; s3, obtaining all target animal information in the video based on a target detection model of deep learning; s4, obtaining the trajectory of a single individual in the animal group through trajectory tracking; s5, performing moving average filtering processing on the track; s6, generating a track of a single individual in the animal group, and generating a tracking video marked with an animal group identity label. The method is high in robustness to environmental noise, can be applied to tracking of multiple animal groups, is high in anti-interference capability, and can solve complex interaction modes such as mutual touch of individuals and mutual intersection of motion paths in the animal groups.

Description

technical field [0001] The invention relates to the technical field of animal behavior, in particular to a method and a system for tracking the trajectory of an animal group. Background technique [0002] During social behavior, animals make many massive and rapid behavioral changes, display rich behavioral patterns, and integrate various information from their own motivations and emotions, as well as the current environment. Animal behavior and its social interaction research can be used to build cognitive and emotional models of normal and diseased animals. It has a wide range of applications in learning and memory research, physiological mechanism research of disease models, and drug evaluation tests. To observe cognitive and social deficits in autism and dementia. [0003] A key challenge in conducting long-term social experiments with multiple animal groups is the ability to obtain reliable animal trajectories. Therefore, how to extract detailed behavioral information ...

AI Technical Summary

Problems solved by technology

However, most of the currently widely used experiments use short-term socialization of a small number of animals, which will miss many important animal social behavior patterns, and the existing human-labeled monitoring videos are not realistic, not only tedious, expensive, but also highly subjective. Difficult to repeat, maintaining automatic tracking and identification of multiple animals in a video sequence is very difficult

In addition, the individuals in the animal group touch each other, the movement paths cross each other, and there will be very complex interaction patterns, such as curling up together, climbing on the back of another animal, etc. In current social experiments, it is usually assumed that the animals keep Visible, non-overlapping, slow motion

Or use other characteristics to mark the identity of animals, such as different colors, sizes, restricted environments (such as a certain animal is restricted to a specific area), etc., but the color marks are easy to be cleaned when the animal is physical. The color mode marking of the traditional color mode mark also has problems such as insufficient repeatability, and the need to retrain the model when using different animal experiments; other markings such as RFID are usually intrusive and will affect or even change the behavior of animals; and the restricted environment is difficult to observe A comprehensive model of animal social behavior

[0004] In addition, in different experimental settings, the color of the animal, the experimental environment, the uniformity of illumination, the imaging quality of the camera, etc., will all affect the reliability of the automatic trajectory tracking of the animal group, while the traditional computer vision system of animal behavior is not suitable for various The degree of self-adaptation is low, the user-friendliness is not strong, and there are often many parameters that need to be adjusted. Experimental settings with low contrast, uneven illumination, dynamic changes in illumination, and dynamic changes in the environment. Individuals in the animal group touch each other, and the movement paths cross each other and interact (such as curling up together, climbing on the back of another animal, etc.) ) is also more sensitive

[0005] The Chinese invention patent "An Animal Individual Recognition System Based on Video Tracking Technology" with the publication number CN109377517A combines the Faster-RCNN multi-target detection model in deep learning with the traditional tracking algorithm Kalman filter to solve multi-target tracking applications Difficulties such as occlusion, track crossing, and poor real-time performance often appear in the video, but its target detection model cannot be adjusted according to the difficulty of video analysis, computational complexity, and analysis time cost. The Euclidean distance between positions is used as the evaluation standard for judging the identity. The detection target position smaller than the threshold is a valid position. This method is very prone to individual animal identity errors when animals contact and cross, and the same target animal is blocked. In the case of reappearance, the target animal will be assigned a different identity before and after disappearing

[0006] Since the existing methods and systems are still difficult to obtain satisfactory results in trajectory tracking, it is necessary to establish a new method for tracking the trajectory of animal groups

Images