Pedestrian detection method combining automatic data augmentation and loss function search

Abstract

The invention belongs to the technical field of pedestrian detection, and particularly relates to a pedestrian detection method combining automatic data augmentation and loss function search, which comprises the following steps: S1-1, establishing a neural network model, a training set and a verification set; S1-2, constructing a search space of an augmentation strategy and a loss function and parameterizing the search space; S1-3, training a neural network model by using a double-layer loop optimization scheme; S2-1, testing the learned neural network model by using the verification set in the step S1-1, calculating the accuracy, entering an application stage if the requirement is met, otherwise, adjusting the neural network model according to an actual result, and returning to the step S1-3; and S3-1, obtaining a pedestrian image from the video sequence, inputting the pedestrian image into the trained neural network model, and accurately positioning the position of the pedestrian by means of the trained neural network model. According to the method, the loss weights of the difficult samples and the simple samples can be adjusted, the optimal combination of the data augmentation strategies can be automatically selected, the balance problem between the data augmentation and the loss function can be effectively processed, and the pedestrian detection accuracy is improved.

Description

technical field [0001] The invention belongs to the technical field of pedestrian detection, and in particular relates to a pedestrian detection method combined with automatic data augmentation and loss function search. Background technique [0002] Pedestrian detection is to use computer vision technology to judge whether there are pedestrians in the image or video sequence and give precise positioning. However, due to the characteristics of both rigid and flexible objects, the appearance of pedestrians is easily affected by clothing, scale, occlusion, posture and viewing angle, etc., so that missed detection and false detection often occur in the detection process. [0003] In the existing pedestrian detection technology, the method of data augmentation is often used to effectively expand the size of the training set, improve the generalization ability of the model, and use image augmentation operations such as cropping, adjusting brightness, and translation to solve the p...

AI Technical Summary

Problems solved by technology

[0005] The invention provides a pedestrian detection method combining automatic data augmentation and loss function search, which can adjust the loss weights of difficult samples and simple samples, automatically select the optimal combination of data augmentation strategies, and can effectively process data augmentation and The balance between loss functions avoids the complicated manual calculation process and design combination, saves time and effort, and at the same time improves the problems of missed detection, false detection and low accuracy caused by environmental factors, and improves detection accuracy

Images