Although the vision transformer has been used in gait recognition, its application in multi-view gait recognition remains limited. Different views significantly affect the accuracy with which the characteristics of gait contour are extracted and identified. To address this issue, this paper proposes a Siamese mobile vision transformer (SMViT). This model not only focuses on the local characteristics of the human gait space, but also considers the characteristics of long-distance attention associations, which can extract multi-dimensional step status characteristics. In addition, it describes how different perspectives affect the gait characteristics and generates reliable features of perspective–relationship factors. The average recognition rate of SMViT for the CASIA B dataset reached 96.4%. The experimental results show that SMViT can attain a state-of-the-art performance when compared to advanced step-recognition models, such as GaitGAN, Multi_view GAN and Posegait.

Infrared point-target detection is one of the key technologies in infrared guidance systems. Due to the long observation distance, the point target is often submerged in the background clutter and large noise in the process of atmospheric transmission and scattering, and the signal-to-noise ratio is low. On the other hand, the target in the image appears in the form of fuzzy points, so that the target has no obvious features and texture information. Therefore, scholars have proposed many object detection methods for dimming infrared images, which has become a hot research topic on account of the flow-rank model based on the image patch. However, the result has a high false alarm rate because the most low-rank models based on the image patch do not consider the spatial-temporal characteristics of the infrared sequences. Therefore, we introduce 3D total variation (3D-TV) to regularize the foreground on account of the non-convex rank approximation minimization method, so as to consider the spatial-temporal continuity of the target and effectively suppress the interference caused by dynamic background and target movement on the foreground extraction. Finally, this paper proposes the minimization of the non-convex spatial-temporal tensor low-rank approximation algorithm (MNSTLA) by studying the related algorithms of the point infrared target detection, and the experimental results show strong robustness and a low false alarm rate for the proposed method compared with other advanced algorithms, such as NARM, RIPT, and WSNMSTIPT.

Infrared thermal imaging technology has been widely applied across multiple sectors of modern society. Nevertheless, constrained by its imaging mechanism and the variability of environmental conditions during imaging, infrared images often suffer from poor quality, characterized by low resolution, weak contrast, blurred edges, missing detail information, and low signal-to-noise ratio. Consequently, enhancing infrared images to boost their quality is essential for facilitating subsequent observation and practical use. In response to this issue, this paper proposes an image enhancement framework integrating brightness adjustment, detail sharpening, and multi-scale fusion, which effectively addresses the common problems of infrared images such as insufficient brightness, low contrast, and blurred details. Initially, the light-map of the original image is extracted and estimated based on the Retinex model, followed by gamma correction to generate the first input for fusion. To offset the detail loss in overexposed regions caused by gamma correction, a sharpened version of the original image is introduced as the second fusion input. Subsequently, weight maps for the two inputs are computed, aggregated, and normalized. Finally, the enhanced image is output through a multi-scale fusion pyramid model. Experiments conducted on MATLAB demonstrate that the enhanced images achieve balanced brightness, along with clear edges and detailed textures.