Multi-scale feature fusion pedestrian detection method based on attention mechanism

Abstract

The invention discloses a multi-scale feature fusion pedestrian detection method based on an attention mechanism, and the method comprises the steps: inputting a training set and a verification set, extracting pedestrian features, and generating a feature map; inputting a network model, training the model; judging whether a specified batch is reached, and if so, outputting the model and verifying the model. According to the multi-scale feature fusion pedestrian detection method based on the attention mechanism, a FCOS algorithm is applied to pedestrian detection, a dense pyramid structure is adopted on the basis of the FCOS algorithm, and top-layer features and bottom-layer features are fused, so that the fused features have bottom-layer space information and top-layer feature detail information, and a pedestrian target can be better identified. And secondly, space and channel attention is fused into the fused features, so that the pedestrian target can be positioned more accurately.

Description

technical field [0001] The invention belongs to the technical field of pedestrian detection, in particular to an attention mechanism-based multi-scale feature fusion pedestrian detection method. Background technique [0002] In the 21st century, China's scientific development has advanced by leaps and bounds, and it has provided considerable convenience for people's daily life. More and more tedious work can be done by computer, instead of every task and job needs to be done by people; in the past, every job was based on manpower, such as farming, which was a waste of time , financial resources will consume a lot of manpower and material resources. However, nowadays, due to the existence of computer big data, while obtaining high-quality results, it also does not occupy public resources and saves a lot of time. Computer vision is an urgent research topic for researchers in the general trend. It applies multiple disciplines such as pattern recognition, image processing and ...

AI Technical Summary

Problems solved by technology

However, in order to achieve better classification results, most of the existing methods regard the development of more complex attention modules as the fundamental task. At the same time, the computational complexity is also rising.

[0009] The pedestrian detection methods proposed above all have certain effects, but when comparing with the real frame of pedestrians, too many anchor frames with aspect ratio are set. In the scene with few pedestrians, such anchor frame mechanism will make the model Generate a large number of anchor boxes that do not contain the target

This will increase the calculation amount of the model and consume a lot of space

Images