Vehicle type identification method and device, electronic equipment and storage medium

Abstract

The invention relates to artificial intelligence, and provides a vehicle type identification method. The method comprises the following steps: carrying out vehicle characteristic extraction on an original image so as to form three characteristic graphs with different scales, improving a conventional single scale characteristic graph into a multi-scale method, thus obviously enlarging recall of vehicles with different scales; executing a box selection operation on the feature maps of the three different scales based on the three sets of anchor boxes so as to generate candidate boxes corresponding to the feature maps of the three different scales, so the problems of missed calling and false detection of the vehicle detection box are effectively relieved; performing blocking operation on complete candidate boxes so as to divide each candidate box in complete candidate boxes into modules with preset sizes; carrying out the pixel point sampling on each module to obtain the output matrix, and adding the partitioning operation, so the pixel deviation problem is effectively relieved, and the positioning precision of the vehicle detection box is improved; therefore, the overall accuracy andrecall rate of vehicle type identification are improved.

Description

technical field [0001] The present invention relates to artificial intelligence, relates to a method for intelligently identifying car models, and in particular to a car model identifying method, device, electronic equipment and computer-readable storage medium. Background technique [0002] The current economy is developing rapidly, and different vehicles are traveling together on the road. Therefore, vehicle model identification is an important solution to the current road traffic safety. [0003] Due to the wide variety of vehicle models, and some models are very similar in appearance or logo, it is difficult to identify these similar models based on the traditional FasterRCNN network, and the accuracy rate is not high, and the car models based on the traditional FasterRCNN The recognition algorithm can achieve a certain accuracy rate in simple scenarios such as sunny days, daytime, and the front of the car, but its accuracy and recall rates are relatively low in difficul...

AI Technical Summary

Problems solved by technology

[0003] Due to the wide variety of vehicle models, and some models are very similar in appearance or logo, it is difficult to identify these similar models based on the traditional FasterRCNN network, and the accuracy rate is not high, and the car models based on the traditional FasterRCNN The recognition algorithm can achieve a certain accuracy rate in simple scenes such as sunny days, daytime, and the front of the car, but its accuracy and recall rate are relatively low in difficult scenes such as haze, rainy days, night, and the side of the car

Images