Pedestrian Attribute Recognition Network and Technology Based on Recurrent Neural Network Attention Model

Abstract

The invention provides a pedestrian attribute recognition network and pedestrian attribute recognition technology based on a cyclic neural network attention model. The pedestrian attribute recognition network includes the first convolutional neural network that uses the pedestrian's original whole-body image as an input to extract the pedestrian's whole-body image features; uses the pedestrian's full-body image features as the first input, and the attention heat map of the attribute group that was concerned at the last moment As the second input, output the attention heat map of the attribute group concerned at the current moment and the cyclic neural network of the pedestrian features that have been partially highlighted; use the pedestrian features that have been partially highlighted as input, and output the attributes of the current attention group A second convolutional neural network for predicting probabilities. The invention utilizes the attention model of the convolutional neural network to mine the association relationship of the spatial position of the attribute area of ​​the pedestrian, more accurately highlight the position of the area corresponding to the attribute in the image, and realize higher recognition accuracy of the attribute of the pedestrian.

Description

technical field

[0001] The invention belongs to the technical field of neural network and image recognition, and in particular relates to a pedestrian attribute recognition network and technology based on a cyclic neural network attention model. Background technique

[0002] Pedestrian attribute recognition technology can help people automatically complete the task of searching for specific people from massive image and video data. However, due to the low image quality of the surveillance video, the small size of the labeled pedestrian attribute data set, and the difficulty in obtaining it, it greatly increases the difficulty of identifying pedestrian attributes from surveillance video images. The existing pedestrian attribute recognition methods based on deep neural network are divided into two categories: convolutional neural network (CNN) method and convolutional neural network combined with recurrent neural network (CNN-RNN). Existing CNN methods such as the DeepMAR met...

Examples

Embodiment 1

[0037] A pedestrian attribute recognition network based on a recurrent neural network attention model, such as figure 1 shown, including:

[0038] Using the original full-body image of the pedestrian as an input to extract the first convolutional neural network of the feature N(x) of the whole-body image of the pedestrian;

[0039] Using the pedestrian full-body image feature N(x) as the first input, the attention heat map A of the attribute group concerned at the last moment t-1 (x) As the second input, output the attention heat map A of the attribute group concerned at the current moment t (x) and pedestrian features H after local highlighting t (x) recurrent neural network;

[0040] Using the partially highlighted pedestrian feature H t (x) As input, a second convolutional neural network that outputs the predicted probability of the attribute of the current group of interest.

[0041] In the pedestrian attribute recognition network provided in this embodiment, the part...

Embodiment 2

[0053] A pedestrian attribute recognition technology based on a recurrent neural network attention model, including:

[0054] S1. Obtain a certain number of pedestrian images with attributes to be identified, and mark whether the images have certain or certain attributes, and obtain a data set that can be used to train the recognition effect of pedestrian attributes; then filter all the attributes marked, Then the filtered attributes are grouped according to the semantic and spatial neighbor relations;

[0055] S2. Using the combination of the Inception network and the convolutional cyclic neural network, construct a pedestrian attribute recognition network based on the convolutional cyclic neural network attention model, including:

[0056] S2-1. Use the Inception network to extract the original full-body image of the pedestrian to obtain the feature N(x) of the whole-body image of the pedestrian;

[0057] S2-2. At time i, use the pedestrian full-body image feature N(x) to c...

technical field

[0001] The invention belongs to the technical field of neural network and image recognition, and in particular relates to a pedestrian attribute recognition network and technology based on a cyclic neural network attention model. Background technique

[0002] Pedestrian attribute recognition technology can help people automatically complete the task of searching for specific people from massive image and video data. However, due to the low image quality of the surveillance video, the small size of the labeled pedestrian attribute data set, and the difficulty in obtaining it, it greatly increases the difficulty of identifying pedestrian attributes from surveillance video images. The existing pedestrian attribute recognition methods based on deep neural network are divided into two categories: convolutional neural network (CNN) method and convolutional neural network combined with recurrent neural network (CNN-RNN). Existing CNN methods such as the DeepMAR met...

Examples

Embodiment 1

[0037] A pedestrian attribute recognition network based on a recurrent neural network attention model, such as figure 1 shown, including:

[0038] Using the original full-body image of the pedestrian as an input to extract the first convolutional neural network of the feature N(x) of the whole-body image of the pedestrian;

[0039] Using the pedestrian full-body image feature N(x) as the first input, the attention heat map A of the attribute group concerned at the last moment t-1 (x) As the second input, output the attention heat map A of the attribute group concerned at the current moment t (x) and pedestrian features H after local highlighting t (x) recurrent neural network;

[0040] Using the partially highlighted pedestrian feature H t (x) As input, a second convolutional neural network that outputs the predicted probability of the attribute of the current group of interest.

[0041] In the pedestrian attribute recognition network provided in this embodiment, the part...

Embodiment 2

[0053] A pedestrian attribute recognition technology based on a recurrent neural network attention model, including:

[0054] S1. Obtain a certain number of pedestrian images with attributes to be identified, and mark whether the images have certain or certain attributes, and obtain a data set that can be used to train the recognition effect of pedestrian attributes; then filter all the attributes marked, Then the filtered attributes are grouped according to the semantic and spatial neighbor relations;

[0055] S2. Using the combination of the Inception network and the convolutional cyclic neural network, construct a pedestrian attribute recognition network based on the convolutional cyclic neural network attention model, including:

[0056] S2-1. Use the Inception network to extract the original full-body image of the pedestrian to obtain the feature N(x) of the whole-body image of the pedestrian;

[0057] S2-2. At time i, use the pedestrian full-body image feature N(x) to c...

Images