A pedestrian appearance attribute identification method based on Inception V3 multi-data-set joint training
An attribute recognition, multi-data technology, applied in image data processing, neural learning methods, character and pattern recognition, etc., can solve problems such as poor generalization ability, and achieve the effect of optimized accuracy and strong generalization ability
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0065] The present invention designs a pedestrian appearance attribute recognition method based on multi-data set joint training of Inception V3, which solves the problem that existing pedestrian appearance attribute recognition methods based on deep learning are vulnerable to illumination, occlusion, target posture changes, and image clarity in monitoring scenes. Influenced by factors and poor generalization ability, accurate identification of pedestrian attributes in the target monitoring scene can be realized, and only a small amount of target scene data is required to achieve it. In particular, the following setting method is adopted: including the following steps:
[0066] 1) Obtain surveillance video clips containing pedestrians, and preprocess pedestrian images;
[0067] 2) Build a new Inception V3 convolutional neural network model;
[0068] 3) Improve the logistic loss loss function;
[0069] 4) Input multiple public datasets for training to obtain a pedestrian appea...
Embodiment 2
[0072] The present embodiment is further optimized on the basis of the above-described embodiments, and further to better realize the present invention, the following setting mode is adopted in particular: the step 1) includes the following specific steps:
[0073] 1.1) Name the captured video frame according to the specified image naming method (for example: 111.jpg, the number represents the image number) and save it to the designated location;
[0074] 1.2) Mark the appearance attributes of all image files to form a data set; the attributes of pedestrians on each pedestrian image are binary attributes, if they have this attribute, the corresponding label value is 1; if not, then The label value is 0. For example, if the pedestrian in the picture wears glasses, the corresponding label value of wearing glasses is 1;
[0075] 1.3) Divide the data set into two parts, namely the training set and the verification set. The training set is used to train the model, and the verificat...
Embodiment 3
[0077] This embodiment is further optimized on the basis of any of the above-mentioned embodiments. Further, in order to better realize the present invention, the following setting method is adopted in particular: the new Inception V3 convolutional neural network model includes 5 convolutional layers , 11 block structures and 4 parallel fully connected layers; the convolutional layer and block structure are used to automatically extract pedestrian attribute features; the fully connected layer is used to combine attribute features to obtain corresponding attribute scores.
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More