Passenger portrait-based face recognition method and device

Abstract

The invention discloses a passenger portrait-based face recognition method. The method comprises the following steps: data acquisition: acquiring passenger travel data of all bus routes in a certain region within a preset time period; constructing a passenger portrait: constructing a multi-dimensional user portrait of each passenger according to the travel data information; face base library construction: according to the travel data information, extracting passenger face images of each bus route, and building a face base library corresponding to each bus route; according to the passenger portraits, extracting frequency information of passengers taking each bus route, setting the passengers with the taking frequency reaching N times as frequent passengers, wherein the face images of the frequent passengers in each bus route serve as the face base library of the bus route; and passenger face recognition: when the passenger takes a bus on site, comparing the passenger with the face in the face base library to complete face recognition.

Description

technical field [0001] The invention relates to the field of artificial intelligence, in particular to a face recognition method and device based on passenger portraits. Background technique [0002] In the payment scenario, face payment requires an extremely low false recognition rate (FAR), and the one-time recognition pass rate must be above 99%. In terms of recognition time, the time-consuming of the overall recognition process should be as small as possible (less than 1 second ), and it is necessary to reduce the probability of the user's second confirmation in order to effectively improve the payment user experience. However, in the actual payment scenario, due to the complexity of the actual situation, such as lighting and angle problems, the bottom database photos are not updated in time, and there are differences with the actual photos of the person, the accuracy of face recognition is often low, reaching In addition, due to the large face base database, the face s...

AI Technical Summary

Problems solved by technology

However, in the actual payment scenario, due to the complexity of the actual situation, such as lighting and angle problems, the bottom database photos are not updated in time, and there are differences with the actual photos of the person, the accuracy of face recognition is often low, reaching In addition, due to the large face base database, the face search and comparison calculation similarity (1:N) in the 100,000-level and one-million-level face base database, the recognition speed and recognition accuracy of face recognition will be reduced, making the face payment experience poor, and even unable to meet the payment misrecognition rate requirements

[0003] In the bus face payment scenario, because the on-site photos are affected by factors such as the surrounding environment, light, and shooting angles, when performing face 1:N comparisons, the number of face photos in the bottom library whose similarity is above a certain threshold is often There are a few to a dozen, and some of them are often very close in similarity. At this time, there will be a problem of misidentification.

Images