Multi-Object Tracking (MOT) technology is dedicated to continuously tracking multiple targets of interest in a sequence of images and accurately identifying their specific positions at different times. This technology is crucial in key application areas such as autonomous driving and security surveillance. However, the application process often requires the coordination of cameras from multiple angles for tracking. Directly studying Multi-Object Tracking algorithms in panoramic scenes is an effective way to address this issue. The uniqueness of panoramic scenes causes target position changes at the boundaries and tracking difficulties due to continuous changes in target scales. To ensure the accuracy of target tracking, this study explores a detection-based tracking method using the newly improved YOLOx detector and the adjusted DeepSORT algorithm. Firstly, YOLOx_s was chosen as the detector because its simple network structure ensures a fast computational speed. During the feature extraction stage, we used the Polarized Self-Attention (PSA) mechanism to capture more feature information, thereby improving the tracking performance on small-scale targets. Secondly, the tracker was improved by adding a camera motion compensation module before predicting the target’s position to mitigate the impact of camera shake on tracking. Finally, to address the difficulty of continuously tracking targets in specific areas of panoramic scenes, this study proposes specific tracking strategies. These strategies effectively resolve the problem of tracking failure caused by target position changes at the boundaries. Experimental results show that improved algorithms have a superior performance on multiple evaluation metrics compared to other algorithms in the field. Compared to the original algorithm, the improved algorithm exhibits a 6% increase in the quantitative metric MOTA, a 7% increase in IDF1, and a 40% decrease in IDSWs, demonstrating its leading performance.

In this paper, we first propose an accurate edge detector using a distance field-based convolutional neural network (DF-CNN). In recent years, CNNs have been proved to be effective in image processing and computer vision. As edge detection is a fundamental problem among them, we try to improve the accuracy of edge detection based on the deep learning framework. The proposed network combines a feature extraction backbone that can fully exploit the multiscale and multilevel information of the edge with the supervised training of the distance field branch to realize the accurate end-to-end object edge detection. The distance field branch is applied to predict the Euclidean distance from nonedge points to the nearest edge point in the feature maps. And the distance information embedded in the predicted distance field map can effectively improve the accuracy of edge detection. The network is trained to minimize the weighted sum of the distance field branch loss and the cross-entropy loss. Our experimental results show that the proposed edge detector achieves better performance than previous approaches and the effectiveness of the proposed distance field branch.

Histogram modification is an important technique for contrast enhancement. Most changes of histogram are based on global or local region grey‐levels information. In this study, a novel grey‐level co‐occurrence matrix (GCOM)‐based histogram equalisation (COHE) method is proposed. A GCOM is a matrix or distribution of co‐occurring grey‐levels at a given offset, in which each row or column vector is actually a conditional histogram. The procedure of COHE has two steps. First, it is to equalise the modified conditional histograms, which are weighted sums of uniformly distributed histograms and the conditional histograms. An adjusting method of weight parameter is also presented in this study. Conditional histograms equalisations have the advantage of enlarging the difference between given grey‐levels and other spatially adjacent grey‐levels. Second, COHE algorithm finds mapping to obtain global enhance by weighting all the conditional translated grey‐levels with original image histogram. However, it could produce over‐enhanced unnatural looking images because of spikes of conditional histogram and original histogram. To deal with this, this study introduces methods of adjusting the conditional histogram and original histogram based on GCOM. Experimental results demonstrate that the proposed method can enhance the images effectively.

Knowledge base question answering (KBQA) aims to provide answers to natural language questions from information in the knowledge base. Although many methods perform well when dealing with simple questions, there are still two challenges for complex questions: huge search space and information missing from the query graphs’ structure. To solve these problems, we propose a novel KBQA method based on a graph convolutional network and optimized search space. When generating the query graph, we rank the query graphs by both their semantic and structural similarities with the question. Then, we just use the top k for the next step. In this process, we specifically extract the structure information of the query graphs by a graph convolutional network while extracting semantic information by a pre-trained model. Thus, we can enhance the method’s ability to understand complex questions. We also introduce a constraint function to optimize the search space. Furthermore, we use the beam search algorithm to reduce the search space further. Experiments on the WebQuestionsSP dataset demonstrate that our method outperforms some baseline methods, showing that the structural information of the query graph has a significant impact on the KBQA task.