Labeling method, device, system and storage medium for pedestrian detection false alarm data

Abstract

The present invention provides a labeling method, device, system and storage medium for false alarm data of pedestrian detection. The method includes: acquiring scene data; performing pedestrian detection on the scene data to obtain a pedestrian prediction frame; and The data marked by the prediction frame at least extract false positive samples of pedestrian detection. The method, device, system, and storage medium for labeling pedestrian detection false positive data according to the embodiments of the present invention can realize automatic labeling of pedestrian detection false positive data, without manpower for labeling pedestrian detection false positive data, and can quickly obtain false pedestrian detection and rejection The training data of the model can improve the training efficiency of the pedestrian false positive rejection model, thereby improving the reliability of pedestrian detection.

Description

technical field [0001] The present invention relates to the technical field of pedestrian detection, and more particularly to a method, device, system and storage medium for labeling false positive data of pedestrian detection. Background technique [0002] Pedestrian detection is an important entry-level technology in video structuring, human posture analysis, and human-computer interaction. Any application related to people will first use pedestrian detection technology to determine the position of the human body. However, the existing pedestrian detection models still have big flaws, and often "falsely report" some objects that look like people. Therefore, in the process of productization, a small and fast pedestrian false positive rejection model is often needed to polish the performance of the algorithm. . [0003] However, since the pedestrian false positive removal model needs to fit the distribution of the pedestrian detection model, the updated version of the pedes...

AI Technical Summary

Problems solved by technology

However, the existing pedestrian detection models still have big flaws, and often "falsely report" some objects that look like people. Therefore, in the process of productization, a small and fast pedestrian false positive rejection model is often needed to polish the performance of the algorithm.

[0003] However, since the pedestrian false positive removal model needs to fit the distribution of the pedestrian detection model, the updated version of the pedestrian detector requires re-labeling data and training models

Model training can be accelerated by expanding the machine. Labeled data is limited by labor costs, making it difficult to grow in scale, which has caused great trouble for product updates.

Images