Search

Recent advances in cross-modal 3D object detection rely heavily on anchor-based methods, and however, intractable anchor parameter tuning and computationally expensive postprocessing severely impede an embedded system application, such as autonomous driving. In this work, we develop an anchor-free architecture for efficient camera-light detection and ranging (LiDAR) 3D object detection. To highlight the effect of foreground information from different modalities, we propose a dynamic fusion module (DFM) to adaptively interact images with point features via learnable filters. In addition, the 3D distance intersection-over-union (3D-DIoU) loss is explicitly formulated as a supervision signal for 3D-oriented box regression and optimization. We integrate these components into an end-to-end multimodal 3D detector termed 3D-DFM. Comprehensive experimental results on the widely used KITTI dataset demonstrate the superiority and universality of 3D-DFM architecture, with competitive detection accuracy and real-time inference speed. To the best of our knowledge, this is the first work that incorporates an anchor-free pipeline with multimodal 3D object detection.

Prior convolution-based road crack detectors typically learn more abstract visual representation with increasing receptive field via an encoder-decoder architecture. Despite the promising accuracy, progressive spatial resolution reduction causes semantic feature blurring, leading to coarse and incontiguous distress detection. To these ends, an alternative sequence-to-sequence perspective with a transformer network termed TransCrack is introduced for road crack detection. Specifically, an image is decomposed into a grid of fixed-size crack patches, which is flattened with position embedding into a sequence. We further propose a pure transformer-based encoder with multi-head reduced self-attention modules and feed-forward networks for explicitly modelling long-range dependencies from the sequential input in a global receptive field. More importantly, a simple decoder with cross-layer aggregation architecture is developed to incorporate global with local attentions across different regions for detailed feature recovery and pixel-wise crack mask prediction. Empirical studies are conducted on three publicly available damage detection benchmarks. The proposed TransCrack achieves a state-of-the-art performance over all counterparts by a substantialmargin, and qualitative results further demonstrate its superiority in contiguous crack recognition and fine-grained profile extraction. This article is part of the theme issue 'Artificial intelligence in failure analysis of transportation infrastructure and materials'.

The management of varying traffic flows essentially depends on signal controls at intersections. However, design an optimal control that considers the dynamic nature of a traffic network and coordinates all intersections simultaneously in a centralized manner is computationally challenging. Inspired by the stable gene expressions of Escherichia coli in response to environmental changes, we explore the robustness and adaptability performance of signalized intersections by incorporating a biological mechanism in their control policies, specifically, the evolution of each intersection is induced by the dynamics governing an adaptive attractor selection in cells. We employ a mathematical model to capture such biological attractor selection and derive a generic, adaptive and distributed control algorithm which is capable of dynamically adapting signal operations for the entire dynamical traffic network. We show that the proposed scheme based on attractor selection can not only promote the balance of traffic loads on each link of the network but also allows the global network to accommodate dynamical traffic demands. Our work demonstrates the potential of bio-inspired intelligence emerging from cells and provides a deep understanding of adaptive attractor selection-based control formation that is useful to support the designs of adaptive optimization and control in other domains.

Objective: This study aims to investigate the crossing behavior of straight-moving drivers when they encounter other straight-moving drivers at unsignalized intersections in China., Background: In China, when two vehicle drivers encounter at an unsignalized intersection, neither driver completely stops his or her vehicle in most cases. Instead, one driver gradually approaches the intersection and dynamically decides to either yield or preempt by gaming with the other vehicle. This process increases the probability of accidents., Method: A total of 305 crossing cases were collected at an unsignalized intersection in Kunming City. Motion parameters were extracted from the video detection program designed by our research group. Based on a logistic regression method, we analyzed decision making moment of straight-moving drivers under crossing conditions, established crossing behavior models, and identified the main factors that affected drivers' decisions. In all, 68 cases observed at a separate intersection were used to validate the established models., Results: For crossing processes at unsignalized intersections in China, straight-moving drivers from the right side completed preemptive/yielding decisions at 1.3 s before reaching the crossing point. However, the majority of straight-moving drivers from the left side completed decisions at 1.1 s before reaching the crossing point. The most important parameter that influenced the drivers' decisions was the difference between the speeds of the two vehicles., Conclusion: An effective method for preventing traffic conflict between two straight-moving drivers is to control the speeds of the vehicles before they enter the intersection., Application: This study explores crossing behavior of straight-moving drivers and provides significant insights for controlling driver behavior., (© 2015, Human Factors and Ergonomics Society.)