Answers are grounded in this paper and cite their sources.
During vehicle operation, occlusion phenomena occur, and the sensing device acquires incomplete image information, posing a new challenge to the recognition of vehicle information. Based on the convolutional and anti-convolutional structure of deep learning, we designed a contextual self-encoder to propose a vehicle de-occlusion recognition method based on contextual information. We constructed a vehicle dataset with 11 categories. Occlusions were generated using the PIL library, and the occlusion region was removed with a contextual self-encoder, improving the recall from 45.16% to 86.25%. To test the generalization performance of the algorithm, we designed experiments to analyze the effects of special weather, occlusion area, and occlusion location on the algorithm's generalization performance, respectively. After de-occlusion, the recall improved from 18.44% to 75.31% in special weather; increasing the occlusion area improved the recall from 33.59% to 85.63%; changing the occlusion location improved the recall from 44.38% to 86.56%. The experiments show that the method has good generalization performance and provides technical support for recognizing obscured vehicles.
No takes yet. Share an insight, caveat, or question.
Xu et al. (2024) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: