The invention discloses an intelligent snapshot system and method based on an FPGA, relates to the technical field of image processing, and relates to a CCM, a board card, an ISP chip deployed on the board card, an FPGA chip and a DDR4. The CCM collects image data of the vehicle and the surrounding environment and transmits the image data to the ISP chip deployed on the board card; the ISP chip performs optimization processing on the image data acquired by the CCM; the FPGA chip is integrated with a Cortex-A53 processor, a model file Loadable is deployed in the Cortex-A53 processor, and the model file Loadable is obtained by performing TensorRT quantization and Compiler compiling on a model obtained by training a Caffe deep learning framework and a ResNet-18 neural network architecture on the basis of violation pictures; and the FPGA chip reads the optimized image data, carries out feature extraction on the read image data, and transmits the extracted feature data to the Cortex-A53 processor, and a model file Loadable of the Cortex-A53 processor carries out image recognition according to the feature data, thereby realizing snapshot of violation behaviors. According to the invention, violation behaviors can be rapidly identified, and timely snapshot and storage of the violation behaviors can be realized.
本发明公开一种基于FPGA的智能抓拍系统及方法,涉及图像处理技术领域,其实现涉及:CCM,板卡,部署于板卡的ISP芯片、FPGA芯片和DDR4;CCM采集车辆与周边环境的图像数据,传输到板卡上部署的ISP芯片;ISP芯片对CCM采集的图像数据进行优化处理;FPGA芯片集成有Cortex‑A53处理器,Cortex‑A53处理器中部署有模型文件Loadable,模型文件Loadable是Caffe深度学习框架和ResNet‑18神经网络架构基于违章图片训练得到的模型经TensorRT量化、Compiler编译得到的;FPGA芯片读取优化处理后的图像数据,对读取的图像数据进行特征提取,并将提取的特征数据传送至Cortex‑A53处理器,Cortex‑A53处理器的模型文件Loadable根据特征数据进行图像识别,实现违章行为的抓拍。本发明可以快速识别违章行为,实现违章行为的及时抓拍和存储。
Intelligent snapshot system and method based on FPGA
一种基于FPGA的智能抓拍系统及方法
2023-05-09
Patent
Electronic Resource
Chinese
License plate snapshot intelligent illumination method and system
European Patent Office | 2020
|