Biometric system identifies individuals by utilizing their behavioral or physical traits. A multimodal biometric system is one of the best choices to overcome the numerous problems including vulnerability to environmental impacts, sensitivity to spoof attacks, non-universality, and intra-class variations problems associated with unimodal biometric system. Hand based biometrics is a swiftly growing field of research. This paper proposes a multimodal biometric system based on hand modalities, which combines the extracted information of dorsal finger crease and finger knuckle print. Dorsal finger crease with a small image dimension is a newly explored biometric trait with a highly distinctive pattern and can be used as a reliable identifier in biometric system. To the best of our knowledge, it is the first time to integrate the dorsal finger crease with finger knuckle print for more secure and better recognition performance of the proposed multimodal biometric system. Additionally, we developed a dorsal finger image acquisition device to construct a novel dataset, USM-FCKP, which simultaneously captures dorsal finger crease and finger knuckle print images. For feature extraction, an efficient texture descriptor, Circular Shift Combination Local Binary Pattern (CSC-LBP) is applied, and compared its performance with other LBP variants, including Basic-LBP, Uniform-LBP, Orthogonal Difference-LBP, and Symmetric Inline Matrix-LBP. Furthermore, feature-level fusion is performed, and a Support Vector Machine (SVM) classifier is employed to evaluate the effectiveness of the proposed dataset and multimodal biometric system. Experimental results demonstrate that the proposed multimodal approach achieves superior recognition performance compared to unimodal methods.

In order to solve the problems of low efficiency and low accuracy of traditional manual detection of road cracks, this paper proposes a vehicle-mounted road crack recognition system based on machine vision and deep learning. It mainly includes three parts: pavement collection and sample set preparation, crack identification model training and crack detection. First, the road surface images are collected through the dual cameras on the vehicle, and image stitching and image enhancement techniques are used for preprocessing to reduce the effects of lighting, vibration, etc. The real road images are combined with the generative adversarial network (GAN) to expand and enrich the sample library, so as to build the customized sample set required for this project. Then, the U-Net convolutional neural network can exclude the non-road parts such as the front vehicle and green belt in the image, and only retain the road surface image, and the model is trained by the YOLOv8 target detection algorithm to obtain the road surface crack recognition model. Finally, the system can detect road surface cracks during vehicle driving, and accurately locate the crack position through the vehicle-mounted satellite positioning system, providing a scientific basis for road maintenance and quality assessment. The system not only improves detection efficiency, but also helps optimize the scheduling of road maintenance resources, and has significant social and economic value.

Falls represent a critical health hazard for the elderly demographic, often inducing severe physical trauma and a subsequent decline in psychological well-being that severely compromises their quality of life. To mitigate these risks and foster independent living, the development of a highly dependable and accurate automated fall detection system is of paramount importance. This paper introduces a machine learning-driven fall detection system that utilizes accelerometer data to categorize various activities. The system was trained and evaluated using three machine learning algorithms: Support Vector Machine (SVM), k-Nearest Neighbors (kNN), and Random Forest (RF), employing the publically accessible SisFall dataset. The method employs a low-pass filter in the data preprocessing phase to eliminate noise from sensor readings, subsequently utilizing an overlapping sliding window technique for feature extraction. The findings indicate that all three classifiers attained outstanding performance, with accuracy, sensitivity, and specificity surpassing 90.0%. Of the three algorithms evaluated, Random Forest exhibited the highest performance, rendering it the optimal selection for fall detection systems intended for elderly individuals.