Method and apparatus for constructing and retrieving a template library
By using a generative variational autoencoder to obtain the mean of vehicle X-ray images as a latent vector, constructing a template library, and retrieving it using Euclidean distance, the problem of low efficiency in template library construction and retrieval in existing technologies is solved, achieving efficient and accurate vehicle recognition.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- NUCTECH CO LTD
- Filing Date
- 2023-10-19
- Publication Date
- 2026-07-21
AI Technical Summary
Existing technologies, especially those based on discriminative neural networks, suffer from low retrieval efficiency and high computational resource consumption when building and retrieving template libraries, particularly when the storage volume is huge. This is especially true in vehicle security inspections, where efficient construction and retrieval of vehicle template libraries are needed to improve recognition accuracy.
A generative variational autoencoder (VAE) model is adopted. By directly obtaining the mean of the template image as the latent vector, a template library is constructed, and Euclidean distance is used for retrieval. This avoids the generation of new samples and the distribution optimization process, and simplifies the retrieval steps.
It improves the efficiency of template library construction and retrieval, reduces the consumption of computing resources, obtains more accurate template data, and enhances the accuracy and robustness of recognition, making it suitable for vehicle recognition in vehicle security inspections.
Smart Images

Figure CN118035476B_ABST