Weak supervision pedestrian detection method and system, medium, equipment and processing terminal

Abstract

The invention belongs to the technical field of pedestrian detection, and discloses a weak supervision pedestrian detection method and system, a medium, equipment and a processing terminal, and the method comprises the steps: constructing a gain operator to carry out high-frequency gain-based saliency contrast enhancement on an infrared image, and obtaining a high-contrast infrared image with highlighted pedestrian information; performing multi-scale decomposition on the image by adopting guided filtering to obtain a sub-base layer and a sub-detail layer of the enhanced infrared image and the visible light image; constructing an objective function based on image feature similarity and image edge reservation, and introducing a marine predator optimization algorithm to generate a fusion image; a brightness perception classifier is introduced to realize data set labeling migration based on a weak supervised learning framework; a convolution block attention model is introduced into the YOLOv5 network, and channel attention and space attention are combined to realize detection of weak supervision pedestrians. According to the method, the importance of pedestrian targets can be improved, background interference can be suppressed, accurate detection of multi-scale targets is realized, and the workload of manual labeling is reduced.

Description

technical field [0001] The invention belongs to the technical field of pedestrian detection, and in particular relates to a weakly supervised pedestrian detection method, system, medium, equipment and processing terminal. Background technique [0002] At present, pedestrian detection is a hot and difficult research in the field of computer vision, and it is widely used in intelligent traffic monitoring, automatic driving, pedestrian behavior analysis, etc. With the help of computer vision technology, it is possible to accurately determine whether an image or video contains a pedestrian, and mark the exact location of the pedestrian. Accurate detection and recognition of pedestrian objects play a very important role in image processing tasks. Visible light vision sensors can collect road scene images with rich texture information and clear features, and are widely used in vehicle-mounted pedestrian detection systems. However, visible light image pedestrian detection systems...

AI Technical Summary

Problems solved by technology

The third type of weakly supervised learning is inaccurate supervision, that is, the given label information is not always true and accurate

[0008] (1) The existing visible light image pedestrian detection system is vulnerable to changes in the lighting environment, especially at night, under heavy rain or fog conditions, pedestrian targets are easy to lose, which can easily cause serious safety accidents

[0009] (2) Existing pedestrian monitoring systems with different sensors still complete the task of pedestrian detection independently, and the final detection result does not use the complementary information between the two sensors, which will lead to the final detection accuracy is usually limited, and largely depends on on the respective imaging system

[0010] (3) The existing manual labeling based on supervised learning has a large workload and the labeling affected by the complex background is inaccurate; in addition, due to the lack of guidance of labeled data, pedestrian detection methods based on unsupervised learning also have unsatisfactory detection accuracy

[0011] (4) The performance of existing methods depends on the robustness and integrity of artificially designed features, and cannot automatically obtain target category attribute information, and with the improvement of detection accuracy requirements, the complexity of artificially extracted features is getting higher and higher

[0012] (5) In most pedestrian detection methods, datasets are usually experimented with a single visible light image or infrared image, which does not actually consider the accuracy and safety of the detection network

[0013] (6) The deep learning method requires a large number of samples and effective labeling of the data, but manual labeling of the samples is time-consuming and laborious in practice

[0014] (7) The third type of weakly supervised learning is inaccurate supervision, that is, the given labeling information is not always true and accurate. The reason may be that the level of the labeler is limited, the labeling process is careless, or the labeling is difficult.

However, pedestrian targets are easily lost in complex environments, which can easily cause serious safety accidents

At the same time, the workload of manual labeling is heavy and the complex background can easily lead to inaccurate labeling

Images