A Crowd Counting Method Based on Small Sample Learning

Abstract

The invention discloses a crowd counting method based on small-sample learning, which is characterized in that it includes the following steps: S1: Aiming at the characteristics of crowd image data, a KNN-based density map generation method is used to preprocess the crowd image and generate a corresponding Density map; S2: Transfer learning is used to transfer features across data; the first few layers of low-level features in crowd images are extracted by a fixed model to retain the knowledge learned in the source domain, and the latter few layers are fine-tuned to make the model suitable for the target domain , the fixed model integrates the knowledge of the source domain and the target domain; S3: Build an adaptive adversarial network counting model, for the input of different resolution images, adaptively learn the fusion scale and aggregate multiple abstraction levels to obtain the final density map. The present invention combines small-sample learning, migration learning and confrontation network, greatly improves the counting performance on the basis of reducing the workload, enhances the robustness of the model, and further improves the counting accuracy.

Description

technical field [0001] The invention relates to the technical field of static image recognition in computer vision, in particular to a crowd counting method based on small sample learning for different resolutions. Background technique [0002] With the rapid development of society and economy, the number of group activities around the world has increased dramatically, and the size of the crowd has become larger and larger. Crowds gather in limited areas, and crowding is common, such as in subways or certain tourist attractions. In this case, overcrowding can lead to traffic delays, accidents or even a serious stampede. In recent years, the frequent stampede incidents at large-scale events at home and abroad have caused a lot of casualties, and the frequent occurrence of such incidents has attracted attention from all sides. [0003] In order to avoid large-scale stampede events in the future as much as possible, resulting in more loss of life and property, crowd image ana...

AI Technical Summary

Problems solved by technology

[0009] The purpose of the present invention is to provide a crowd counting method based on small-sample learning in view of the above-mentioned problems, and to solve many complex problems such as noise interference and scarce data in the application of actual crowd scenes, and combine small-sample learning and migration The solution strategy of learning and confronting the network; on the basis of reducing the workload, the counting performance is greatly improved, the robustness of the model is enhanced, and the counting accuracy is further improved

Images