A Human Gait Recognition Model Based On The Combination Of KNN And CNN

Gait recognition based on video data can effectively overcome the limitations of common identification methods(such as face recognition,fingerprint recognition and iris recognition),and has been widely used in many scenes such as criminal investigation,smart medical treatment,smart home and smart factory.Due to the complexity of practical application scenarios,such as heavy clothing such as coats or cotton jackets,the existing gait recognition methods still have the problem of low accuracy.Based on KNN and CNN,this paper conducts research on gait recognition in complex scenes.The main results are as follows:Firstly,this paper organically integrates the advantages of SOLOv2 instance segmentation method and KNN image matting method,and designs a gait contour extraction algorithm with more obvious features.Numerical experiments show that compared with the existing methods,the contour images extracted by the new method have more distinct edges and angles at the edges and joints,and have a higher coincidence with the original RGB images.Then,the Gait Set model is improved to extract local features more effectively.On this basis,a Gait Set gait recognition algorithm was designed,named SKMGL-Gait Set(SOLOv2 KNN Matting Global Local Gait Set),which organically combined local and global features.Finally,the numerical comparison experiment based on CASIA-B data set shows that the recognition accuracy of SKMGL-Gait Set algorithm is significantly improved compared with the existing method.For example,in backpack and overcoat,the accuracy of the new algorithm is improved by 2.55% and 8.69% respectively compared with the original Gait Set algorithm.