No abstract

Robust driver attention prediction for critical situations is a challenging computer vision problem, yet essential for autonomous driving. Because critical driving moments are so rare, collecting enough data for these situations is difficult with the conventional in-car data collection protocol-tracking eye movements during driving. Here, we first propose a new in-lab driver attention collection protocol and introduce a new driver attention dataset, Berkeley DeepDrive Attention (BDD-A) dataset, which is built upon braking event videos selected from a large-scale, crowd-sourced driving video dataset. We further propose Human Weighted Sampling (HWS) method, which uses human gaze behavior to identify crucial frames of a driving dataset and weights them heavily during model training. With our dataset and HWS, we built a driver attention prediction model that outperforms the state-of-the-art and demonstrates sophisticated behaviors, like attending to crossing pedestrians but not giving false alarms to pedestrians safely walking on the sidewalk. Its prediction results are nearly indistinguishable from ground-truth to humans. Although only being trained with our in-lab attention data, the model also predicts in-car driver attention data of routine driving with state-of-the-art accuracy. This result not only demonstrates the performance of our model but also proves the validity and usefulness of our dataset and data collection protocol.

The perception of attractiveness is essential for choices of food, object, and mate preference. Like perception of other visual features, perception of attractiveness is stable despite constant changes of image properties due to factors like occlusion, visual noise, and eye movements. Recent results demonstrate that perception of low-level stimulus features and even more complex attributes like human identity are biased towards recent percepts. This effect is often called serial dependence. Some recent studies have suggested that serial dependence also exists for perceived facial attractiveness, though there is also concern that the reported effects are due to response bias. Here we used an attractiveness-rating task to test the existence of serial dependence in perceived facial attractiveness. Our results demonstrate that perceived face attractiveness was pulled by the attractiveness level of facial images encountered up to 6 s prior. This effect was not due to response bias and did not rely on the previous motor response. This perceptual pull increased as the difference in attractiveness between previous and current stimuli increased. Our results reconcile previously conflicting findings and extend previous work, demonstrating that sequential dependence in perception operates across different levels of visual analysis, even at the highest levels of perceptual interpretation.

In the application of structural health monitoring, the measured data might be temporarily or permanently lost due to sensor fault or transmission failure. The measured data with a high data loss ratio undermine its ability for modal identifications and structural condition evaluations. To reconstruct the lost data in the field of structural health monitoring, this study proposes a deep convolutional generative adversarial network which includes a generator with encoder–decoder structure and an adversarial discriminator. The proposed generative adversarial network model needs to understand the content of the complete signals, as well as produce realistic hypotheses for the lost signals. Given the data stably measured before the occurrence of data loss, the generator is trained to extract the features maintained in the data set and reconstruct lost signals using the responses of the remaining functional sensors alone. The discriminator feeds back the distinguished results to the generator to improve its reconstruction accuracy. When training the model, the reconstruction loss and the adversarial loss are employed to better handle the low-frequency features and high-frequency features of the signals. The effectiveness and efficiency of the proposed method are validated by two case studies. As the number of training epoch increases, the reconstructed signals learn the features from low-frequency to high-frequency, and the amplitude of the reconstructed signals gradually increases. It can be seen that the final reconstruction signals match well with the real signals in the time domain and frequency domain. To further demonstrate the applicability of the reconstructed signals in data analysis, the reconstructed acceleration data are used to accurately identify the modal parameters in the numerical case, and the vehicle-induced responses are precisely decomposed from the reconstructed strain data in the field case. Finally, the reconstruction capacity is also investigated with the different numbers of the faulted strain gauges.

Collecting the information of traffic load, especially heavy trucks, is crucial for bridge statistical analysis, safety evaluation, and maintenance strategies. This paper presents a traffic sensing methodology that combines a deep learning based computer vision technique with the influence line theory. Theoretical background and derivations are introduced from both aspects of structural analysis and computer vision techniques. In addition, to evaluate the effectiveness and accuracy of the proposed traffic sensing method through field tests, a systematic analysis is performed on a continuous box-girder bridge. The obtained results show that the proposed method can automatically identify the vehicle load and speed with promising efficiency and accuracy and most importantly cost-effectiveness. All these features make the proposed methodology a desirable bridge weigh-in-motion system, especially for bridges already equipped with structural health monitoring system.