基 于 改 进 TLD跟踪算法的生猪视频跟踪

Because of its vulnerable to large-scale infection, pig farms are mostly set in the sparsely populated areas. The traditional personnel supervision system has the problems of heavy workload and low efficiency. In order to realize the non-contact supervision of pigs, it was proposed a pig video tracking method based on the improved T L D tracking algorithm with the pig house monitoring video as the data source. The SSD network training model was introduced to train the samples, improve the target detection part of the tracking algorithm, and bring the recognition results directly into the tracking algorithm, which greatly reduced the number of windows in the detection module, and improved the detection accuracy and operation speed. In the test set, the improved model proposed in this paper could recognize the lying, motionless and active behaviors of pigs in the video, with an average accuracy of 94. 26%, 95.67% and 91.36% respectively. Compared with the traditional T L D algorithm, the recognition success rate and accuracy were improved by 10.7% and 6.41 % respectively. The improved T L D tracking algorithm can not only improve the recognition accuracy, but also ensure the recognition efficiency. It can be used to monitor the active information of pigs in the whole time period, and the detection results can be used as one of the references for healthy pig breeding.