Human body recognition model, training method and system thereof, storage medium and equipment

Abstract

The invention discloses a human body recognition model, a training method and system thereof, a storage medium and equipment. The training method of the human body recognition model comprises the following steps: A, screening a training image conforming to a preset standing state from a database as a training sample image, and adjusting the training sample image to a preset size; B, performing overturning processing and rotating processing on the adjusted training sample image; C, randomly selecting a rotation angle from a preset rotation range to rotate the image obtained in the step B; D, the convolutional neural network performing reverse operation corresponding to the rotation processing in the step B on the image obtained in the step C to obtain human body features in the image; E, calculating a loss gradient by utilizing a loss function; updating parameters of the convolutional neural network according to the loss gradient; and F, repeating the steps B-E until the loss function converges. The human body recognition model obtained through the training method of the human body recognition model can have a higher recognition rate for human body recognition under the non-human-body upright conditions such as handstand and body bending.

Description

technical field [0001] The embodiments of the present invention relate to the technical field of human body recognition, and in particular to a human body recognition model and its training method, system, storage medium and equipment. Background technique [0002] Human body recognition has a wide range of applications in video surveillance, video / image search, human body tracking and other fields. The main research in video surveillance is pedestrian re-identification. At present, the most effective pedestrian re-identification algorithm is the recognition algorithm based on deep convolutional neural network, but the pedestrian re-identification algorithm in the prior art only considers the human body in the upright situation. Recognition, the recognition effect of the human body in non-upright postures such as handstand and bending is poor, and it is far from meeting the needs of human body search and human body tracking. Therefore, it is an urgent technical problem to p...

AI Technical Summary

Problems solved by technology

The main research in video surveillance is pedestrian re-identification. At present, the most effective pedestrian re-identification algorithm is the recognition algorithm based on deep convolutional neural network, but the pedestrian re-identification algorithm in the prior art only considers the human body in the upright situation. Recognition, the recognition effect is poor for the non-upright posture of the human body such as handstand and bending, and it is far from meeting the needs of human body search and human body tracking.

Images