Real-time pedestrian clothing color identification method under video

Abstract

The invention provides a real-time pedestrian clothes color identification method under a video, belongs to the field of computer vision and the technical field of image processing, and can quickly extract pedestrians and pedestrian clothes through a Yolo target detection algorithm. The method comprises the following steps: randomly serializing a picture, converting the picture into an element sequence, shielding structural information and spatial information in the picture, and reserving color information of the picture. And according to the conversion of the RGB-HSV color space, encoding values of 16777216 kinds of RGB to form a color dictionary. According to the mapping relation of the color dictionary, training pictures are generated through a random function superposition algorithm, and a large number of training pictures are generated. A Seq2Seq model is constructed, an attention mechanism in a traditional Seq2Seq model is removed, the number of neurons in an output layer of a decoder is reduced, the number of model parameters is reduced, and the operation speed of the model is increased.

Description

technical field [0001] The invention relates to the field of computer vision and the technical field of image processing, in particular to a real-time pedestrian clothing color recognition method under video. Background technique [0002] In various fields involved in computer vision, the accurate identification of pedestrians under surveillance video is particularly important, and clothing color, as an important attribute of pedestrians, is a key element for accurate pedestrian analysis. [0003] However, the current methods for identifying the color of pedestrians’ clothing have the following problems: [0004] (1) Generally, the pedestrians and clothing are cut out from the overall picture through the target detection algorithm or the semantic segmentation algorithm. However, the semantic segmentation algorithm is to classify each pixel in the picture, so the running speed is extremely slow, and it is difficult to meet the speed requirements of real-time interception in ...

AI Technical Summary

Problems solved by technology

However, the semantic segmentation algorithm classifies each pixel in the picture, so the running speed is extremely slow, and it is difficult to meet the speed requirements of real-time interception in video surveillance on most machines, while most target detection algorithms use traditional two-stage target detection Algorithms, such as FAST-RCNN, FASTER-RCNN, although these algorithms have been improved in speed compared with semantic segmentation algorithms, they still cannot meet the speed requirements of real-time interception in video surveillance on some machines

[0005] (2) After the pedestrians are cut out from the picture, a clustering algorithm is usually used to extract the main color, but most of the clustering algorithms can only be deployed and run on the CPU, and GPU computing power cannot be used for calculation, so the speed is relatively slow

[0006] (3) After the pictures of pedestrians are intercepted, the convolutional neural network is usually used for color recognition, but because the convolutional neural network is more sensitive to the structural information and spatial information of the picture; as the network deepens, the underlying color information loss of the picture is relatively large. Large, the ability to learn and understand the color information at the bottom of the picture is weak, so the effect of clothing color recognition with more interference items is poor

Images