Multi-attitude pedestrian detection method and computer storage medium based on deep learning

Abstract

The present invention provides a multi-pose pedestrian detection method based on deep learning and a computer storage medium. The method includes the following steps: S1, defining multiple pedestrian poses, and generating a data set of multi-pose pedestrian targets; S2, dividing the data set according to different pedestrian poses Classify, and divide the data sets of different pedestrian poses into two parts, the training set and the test set; S3, combine the training sets of all pedestrian poses into a total training set for training, and obtain the training model; S4, use the training model to The test sets of different pedestrian poses are tested separately; S5, pedestrian detection is performed according to the test results. According to the method of the embodiment of the present invention, by classifying different postures of pedestrians, different postures of pedestrians can be effectively detected, and the detection accuracy of pedestrians with different postures in complex environments is improved to a certain extent.

Description

technical field [0001] The present invention relates to the field of target detection, and more specifically, to a multi-pose pedestrian detection method based on deep learning and a computer storage medium. Background technique [0002] Pedestrians are non-rigid targets. In reality, there are often multi-modal pedestrian targets in many complex scenes, such as sitting, standing, lying, squatting, etc. Even the same target has different motion postures at different times, and with the Depending on the viewing angle, the shape of pedestrians seen will be different. As a branch of target detection, pedestrian detection is the premise and foundation of pedestrian re-identification and pedestrian tracking. The main task is to detect pedestrians from the input data and determine the position of pedestrians in the input data. It is widely used in intelligent video surveillance. , human-computer interaction, automotive assisted driving and other fields, has good development potent...

AI Technical Summary

Problems solved by technology

[0009] 1. Low resolution: The pedestrian target may be far away from the observation point, and the resolution of the target is very small

[0010] 2. Unstable picture quality: due to lighting, weather, and shooting quality problems, the captured pictures may have problems such as blur, exposure, dimness, noise, etc., resulting in unstable detection results of the algorithm

In the process of detection, we not only need to consider the appearance characteristics of each part of the target, but also consider the positional relationship of each part, which increases the complexity of the detection model

Images