9 December 2022 in Information Fusion
6 January 2024 in Knowledge-Based Systems
14 March 2025 in Engineering Applications of Artificial Intelligence
27 December 2023 in Information Fusion
20 August 2024 in IEEE Transactions on Instrumentation and Measurement
Robust traffic object detection has significant meanings for applications like intelligent transportation and surveillance. However, detection in the open world faces various adverse factors like occlusion, bad weather, and low light. Current endeavors focus on proposing better learnable modules for better feature representation, but their generalization ability is limited by the quality and diversity of training data. In this paper, we propose an IA-Det to achieve robust detection at various challenging transportation scenes by fully exploring semantics and suppressing the noises. First, an Iterative Attention based on t -mixture probability model is proposed to collect rich semantics at the localization branch. Unlike earlier attention methods, the IA module inherits the robust characteristic of t -mixture, and asymmetrically redistributes the attention to features based on their similarity to semantic centers. Therefore, out-of-distribution features can receive greater opportunities for clustering into their semantic neighbors rather than being suppressed as noises. For the classification branch, an effective and efficient Classification Responding Block is designed to activate classification-aware semantics from the collected attention map. Comprehensive experiments on normal, rainy, foggy, and low light fusion datasets show that the proposed IA-Det achieves state-of-the-art performance with an impressive generalization ability.
10 April 2023 in TechRxiv
Image registration is a technique of matching and superimposing of features, which can be better applied to high-level vision tasks. However, high-level vision tasks often pay more attention to analysis of target areas. In summary, we rethink the collaborative relationship between image registration and high-level vision tasks, and propose a local registration strategy for constructing a robust adaptive attention gauge fields, and construct robust adaptive attention gauge fields to register images focus more on important target areas. In order to improve the robustness and timeliness of the algorithm for feature matching of target areas in complex environments, we propose a robust adaptive probability distribution(RA), and construct robust adaptive mixed model expectation maximum attention(RAMM-EMA). In order to make robust adaptive parameters to reach the global optimum in deep learning, we designed a simulated annealing method to optimize the robust adaptive parameters during training. In order to focus on important regions in the image as much as possible, propose a feature fusion map with weak boundaries and probabilistic properties, called Robust Adaptive Attention Gauge Fields(RAA-Gauge Fields). High-level vision tasks experiments on image fusion, object delect and 3D-reconstruction, the method presented in this paper is the most helpful, and registration effect of the target areas perform best under more extreme conditions.