Summary

Recommendation ITU-T F.743.22 specifies requirements and architecture for algorithm-training systems (ATSs) for intelligent video surveillance (IVS) and provides the algorithm-training workflow within it. The intelligent analysis algorithm in IVS needs to train a large amount of actual scene data to improve the accuracy and the recall of identification. The ATS can collect video and image data from the IVS, complete sample data selection and annotate data, as well as training and deploying algorithms in the IVS, so that the IVS has the reasoning capability of new scenarios and the ATS can continuously iteratively improve on algorithm identification performance. Recommendation ITU-T F.743.22 aims to solve the problems of difficult application of algorithm customization in various industries, long algorithm development and iteration cycle, and effective protection of user privacy and improvement of data security.