Filtering method, device and system suitable for target recognition data

Abstract

The invention provides a filtering method, device and system suitable for target recognition data. The method comprises the following steps: obtaining initial target recognition data comprising at least one target, the initial target recognition data being a recognition result outputted after a target detection algorithm recognizes a monitoring video or image; collecting a specified area in the initial target recognition data, and adjusting a target in the specified area based on the type of the specified area to obtain first target recognition data; and inputting the first target recognition data into a confidence coefficient prediction model to obtain confidence coefficients of targets in the first target recognition data, and filtering the targets of which the confidence coefficients are lower than a set threshold value to obtain second target recognition data. According to the scheme, the target recognition data is screened and filtered in combination with a multi-dimensional filtering mechanism, and the recognition accuracy is improved on the basis of not changing a target detection algorithm.

Description

technical field [0001] The present application relates to the field of data processing, in particular to a filtering method, device and system suitable for target recognition data. Background technique [0002] With the continuous optimization of deep learning algorithms, the detection performance of image or video target detection tasks has also been significantly improved, which can replace manual detection to intelligently find specific targets in images or videos. Target detection algorithms are widely used in various fields. For example, they can be used to identify and obtain illegal urban management events in the field of urban management. In specific application scenarios, the recognition results obtained by such target detection algorithms will be used in large Data analysis, however, in order to ensure the accuracy of the final analysis data, it is often necessary to manually verify the recognition results obtained by the target detection algorithm again, and the m...

AI Technical Summary

Problems solved by technology

Target detection algorithms are widely used in various fields. For example, they can be used to identify and obtain illegal urban management events in the field of urban management. In specific application scenarios, the recognition results obtained by such target detection algorithms will be used in large Data analysis, however, in order to ensure the accuracy of the final analysis data, it is often necessary to manually verify the recognition results obtained by the target detection algorithm again, and the manual verification method has fatal problems of heavy workload and low efficiency, and manual verification method cannot guarantee the accuracy of the verification

[0003] Taking the verification requirements of the identification results of urban management violation events as an example, the current identification of urban management violation events has at least the following problems in practical applications: 1. Due to the lack of training samples in some application scenarios, the detection effect of the target recognition algorithm itself Not good, and cannot be directly applied to the detection of violations in actual scenes; 2. Considering that the surveillance cameras used in different regions will affect the content of the surveillance video due to the surrounding environment and surveillance installation angle, which will lead to the detection of the same target The recognition accuracy of the algorithm is different in different scenarios; 3. The current target recognition algorithm has a large number of repetitions or false positives in actual detection, which affects the quality of urban management recognition services

Images