Develop a Lightweight Convolutional Neural Network to Recognize Palms Using 3D Point Clouds

Biometrics has become an important research issue in recent years, and the use of deep learning neural networks made it possible to develop more reliable efficient recognition systems. Palms have been identified as one most promising candidates among various biometrics due their unique features easy accessibility. However, traditional palm methods involve 3D point clouds, which can be complex difficult work with. To mitigate this challenge, paper proposes two are Multi-View Projection (MVP) Light Inverted Residual Block (LIRB).The MVP simulates different angles that observers observe palms reality. It transforms clouds into multiple 2D images effectively reduces loss mapping data data. Therefore, greatly reduce complexity system. In experiments, demonstrated remarkable performance on famous models, such VGG or MobileNetv2, with a particular improvement smaller models. further improve small applies LIRB build lightweight CNN called Tiny-MobileNet (TMBNet).The TMBNet only few convolutional layers but outperforms baselines PointNet PointNet++ FLOPs accuracy. The experimental results show proposed method challenges recognizing through palms. not system also extends CNN. These findings significant implications for developing could lead improvements fields, access control security control.