An intelligent recognition method for the angle position of a car picture

Abstract

The invention discloses an intelligent recognition method for the angular position of a car picture. By acquiring a plurality of car pictures with angle marks and performing preprocessing, a first picture is generated, and grayscale processing is performed on the first picture to obtain a first grayscale picture. Use the SIFT algorithm to extract the feature area of ​​the first grayscale image to generate the first feature area, and perform highlight processing on the first feature area to generate the first feature pixel point, and then substitute the first feature pixel point into the chi-square formula for calculation to obtain the car The most significant first angle position of the image angle classification, input the pixel and angle category of the first angle position to the CNN model, output a 256-dimensional vector, and cluster the 256-dimensional vector through k-means to generate multiple angle categories , and input multiple angle categories and 256-dimensional vectors into the vehicle angle recognition model to generate the optimal angle position, which can reduce the participation of personnel and greatly improve the recognition efficiency and accuracy.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to an intelligent recognition method for the angular position of a car picture. Background technique [0002] The existing object recognition technology of car pictures mainly uses shape features to represent the target object area in the picture, and then uses a classifier to classify the obtained feature vectors, or manually defines the angle position labels of car pictures, or through image synthesis The device acquires images, and then projects the images to grasp the surrounding conditions and the positional relationship of the car relative to other objects, so as to identify the position of the car. [0003] However, in real life, due to the variability of the shape of the car and the interior of the cockpit, the robustness of identifying the position of the car through the contour line features of the object is poor; secondly, for many interferences such as partial o...

AI Technical Summary

Problems solved by technology

[0003] However, in real life, due to the variability of the shape of the car and the interior of the cockpit, it is less robust to identify the position of the car through the contour line features of the object; secondly, for many interferences such as partial occlusion, viewing angle changes, noise and blur Factors are relatively sensitive; and manual labeling requires proofreading, which is time-consuming and labor-intensive; and logical judgments based on projection principles are complex and difficult to grasp, resulting in high costs

Images