When the colour of the target is similar to the colour of the background, or illumination of a scene, scale and shape of the target change significantly, the traditional Kernel-Correlated Filtering (KCF) algorithm often fails to track the target. Therefore, a multi-features fusion target tracking algorithm based on KCF is proposed. The algorithm makes the most of the illumination and rotation invariance of Local binary patterns (LBP) feature, the effective capability of extracting colour and contour information from target by colour Name (CN) feature and Histogram of Oriented Gradient (HOG) feature respectively. The multi-features linear weighted fusion rule is established by calculating the HOG, LBP and CN histograms of the current frame, and calculating the weighted fusion coefficients of the normalized histograms according to the variance of each histogram adaptively. On this basis, multi-scale sampling is used in the target areas and bilinear interpolation is used to adjust the sample size, which can improve the tracking robustness of KCF algorithm to adapt to the target scale. The comparison experimental results of 50 video data sets in OTB-2015 show that the average success rate and precision of the proposed algorithm are 24.74% and 8% higher than that of KCF tracking algorithm respectively, and also significantly better than other tracking algorithms ranked top 10 in OTB-2015, UAV123 and LaSOT data sets respectively.

A tripod-type cold atomic system driven by double-probe fields and a coupling field is explored to store double images based on the electromagnetically induced transparency (EIT). During the storage time, an intensity-dependent signal field is applied further to extend the system with the fifth level involved, then the cross-phase modulation is introduced for coherently manipulating the stored images. Both analytical analysis and numerical simulation clearly demonstrate a tunable phase shift with low nonlinear absorption can be imprinted on the stored images, which effectively can improve the visibility of the reconstructed images. The phase shift and the energy retrieving rate of the probe fields are immune to the coupling intensity and the atomic optical density. The proposed scheme can easily be extended to the simultaneous storage of multiple images. This work may be exploited toward the end of EIT-based multiple-image storage devices for all-optical classical and quantum information processings.

For networks that process 3D data, estimating the orientation and position of 3D objects is a challenging task. This is because the traditional networks are not robust to the rotation of the data, and their internal workings are largely opaque and uninterpretable. To solve this problem, a novel equivariant self-supervised vector network for point clouds is proposed. The network can learn the rotation direction information of the 3D target and estimate the rotational pose change of the target, and the interpretability of the equivariant network is studied using information theory. The utilization of vector neurons within the network lifts the scalar data to vector representations, enabling the network to learn the pose information inherent in the 3D target. The network can perform complex rotation-equivariant tasks after pre-training, and it shows impressive performance in complex tasks like category-level pose change estimation and rotation-equivariant reconstruction. We demonstrate through experiments that our network can accurately detect the orientation and pose change of point clouds and visualize the latent features. Moreover, it performs well in invariant tasks such as classification and category-level segmentation.

Equivariant networks have recently made significant strides in computer vision tasks related to robotic grasping, molecule generation, and 6D pose tracking. In this paper, we explore 3D mesh object analysis based on an equivariant masked autoencoder to reduce the model dependence on large datasets and predict the pose transformation. We employ 3D reconstruction tasks under rotation and masking operations, such as segmentation tasks after rotation, as pretraining to enhance downstream task performance. To mitigate the computational complexity of the algorithm, we first utilize multiple non-overlapping 3D mesh patches with a fixed face size. We then design a rotation-equivariant self-attention mechanism to obtain advanced features. To improve the throughput of the encoder, we design a sparse token merging strategy. Our method achieves comparable performance on equivariant analysis tasks of mesh objects, such as 3D mesh pose transformation estimation, object classification and part segmentation on the ShapeNetCore16, Manifold40, COSEG-aliens, COSEG-vases and Human Body datasets. In the object classification task, we achieve superior performance even when only 10% of the original sample is used. We perform extensive ablation experiments to demonstrate the efficacy of critical design choices in our approach.

In the field of GPU graphics processing, pixel performance is often the bottleneck of system performance. Reducing the load of the pixel engine is of great significance to improving the graphics performance of the chip. After rendering low-resolution images, the resolution is scaled to the desired size, which has become a widely used performance optimization method. Inspired by this approach, this paper proposes a method based on FPGA acceleration to scale 1K low-resolution (LR) images to 4K super-resolution (SR) images. The combination of bicubic interpolation and improved SRCNN is adopted to scale images. The algorithm structure is optimized through High-Level Synthesis (HLS), and the accelerated circuit is deployed on Field Programmable Gate Array (FPGA). Experimental result shows that FPGA achieves image super-resolution with lower latency than Intel Core i5- 8300H CPU and lower power consumption than NVIDIA GTX 2080Ti GPU.