Research And Application Of Face Recognition Algorithm In Vehicular Scenes

With the development and growth of the online car industry,online cars are gradually integrated into the daily way of people’s travel,but the general public is increasingly concerned about the safety of online car travel due to the influence of uneven quality of online car drivers,loopholes in platform gate-keeping and uncontrolled complexity of scenarios.The application of face recognition technology in in-vehicle scenarios can supervise and regulate online taxi drivers.However,most of the use scenes of the two-dimensional face recognition technology that has been more mature are controlled,and there are problems of insufficient security in two-dimensional face recognition,and these problems have become bottlenecks in the development of face recognition.This topic comes from the science and technology project of an enterprise in Guangzhou City,in order to effectively solve the problem of two-dimensional face recognition,this topic uses three-dimensional face recognition technology,and expensive three-dimensional scanning equipment is obviously not suitable for application in the car,so this paper,with the help of low-cost D435 depth camera developed by Intel,around the vehicle scene driver image processing,face detection and face feature extraction and other issues sub module for the research and application of the algorithm.The specific research contents are as follows.(1)Based on the working principle of D435 depth camera,we improve the quality of face data through camera calibration,image alignment,depth image background rejection and depth hole filling processing,so as to do the basic work for subsequent face detection and feature extraction.(2)For the vehicle environment,the traditional Ada Boost algorithm is innovatively improved and combined with the skin color segmentation model and the depth detection module to propose a face detection algorithm that meets the actual application scenarios.The self-built RGB-D face data set is used to analyze the face detection algorithm in this paper.After experimental verification,the combination of the three can efficiently and accurately detect the driver’s face(3)Based on the characteristics of multimodal face recognition,this paper improves Res Net18 by adding an attention mechanism to its residual block and fusing the features of RGB and Depth modalities to design an RGB-D face recognition algorithm,and analyzes the recognition accuracy of this algorithm with a benchmark dataset and a self-built RGB-D face dataset.(4)Combine the image processing technology,face detection algorithm and face recognition algorithm proposed in this paper to build a face recognition system,and verify the face recognition performance of the system in in-vehicle scenarios through real-vehicle experiments.The experimental results prove that the average accuracy of face recognition of the algorithm system proposed in this paper reaches 96.92% in real vehicle verification,with high accuracy and good robustness of face recognition,which can effectively identify the identity of online taxi drivers and play a supervisory role.