Captive livestock multi-target tracking method and device based on neural network

Abstract

The invention discloses a captive livestock multi-target tracking method and device based on a neural network, and the method comprises the following steps: building a multi-target tracking model which comprises a YOLOv3 improved network, a connection transition layer and a Dep Sort improved network, wherein the YOLOv3 improved network adds the obtaining of target feature data, and the Dep Sort improved network omits a neural network for apparent feature extraction; acquiring a plurality of images of target positions of existing captive livestock as a training data set to train the multi-target tracking model; and obtaining a livestock to-be-tracked video, and inputting each frame of image of the video into the trained multi-target tracking model to realize tracking of each livestock target. The method is low in complexity, high in tracking accuracy and high in tracking rate.

Description

technical field [0001] The invention relates to the technical field of computer vision and artificial intelligence, in particular to a neural network-based multi-target tracking method and device for captive livestock. Background technique [0002] Using computer vision technology to manage livestock has gradually become the core technology of artificial intelligence-based animal husbandry. At present, livestock breeding based on computer vision technology can only obtain the location of livestock and identify the species of livestock on the video image, but cannot track every livestock target in the video. In fact, efficient AI livestock management technology needs to continuously capture and collect information on the movement characteristics of livestock, and macroscopic target positioning and species identification cannot pay attention to the health status of individual livestock. The lack of multi-target tracking of livestock AI management brings high livestock managem...

AI Technical Summary

Problems solved by technology

At present, livestock breeding based on computer vision technology can only obtain the location of livestock and identify the species of livestock on the video image, but cannot track every livestock target in the video

In fact, efficient AI livestock management technology needs to continuously capture and collect information on the movement characteristics of livestock, and macroscopic target positioning and species identification cannot pay attention to the health status of individual livestock

The lack of multi-target tracking of livestock AI management brings high livestock management costs, which is not conducive to the realization of low-cost artificial intelligence livestock breeding. Efficient artificial intelligence management needs to monitor the movement and active status of animal targets in the surveillance video. Keep track of

Traditional multi-target tracking algorithms based on image recognition algorithms are lacking in time complexity performance and tracking accuracy performance, making it difficult for real-time video-based multi-target tracking algorithms to meet industrial and practical needs

Images