Traffic light detection system and method based on RAW domain data

By improving the model structure, the problem of low traffic light detection accuracy in the prior art is solved, high-precision detection in complex environments is realized, and computational complexity is reduced.

CN119723508APending Publication Date: 2025-03-28WUHAN HUANYU ZHIXING TECH CO LTD
0 Cites 0 Cited by

Patent Information

Application Number
CN202411460432.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-10-18
Publication Date
2025-03-28

AI Technical Summary

Technical Problem

The existing traffic light detection technology has low detection accuracy under complex traffic environments, lighting changes, weather conditions, occlusion and diversified traffic light styles and layouts. After commonly used RGB images are processed by the camera's built-in ISP, the dynamic range and color range are limited, resulting in a decrease in detection accuracy.

Method used

The traffic light detection system based on RAW domain data is adopted to process the four-channel data of RAW domain data by improving the model structure, and combine the preset backbone network, feature fusion module and detection module to realize high-precision detection of traffic lights.

Benefits of technology

It significantly improves the accuracy and robustness of traffic light detection, can maintain high detection accuracy in special scenarios, and reduces the computational complexity and inference time.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention provides a traffic light detection system and method based on RAW domain data, and relates to the technical field of image processing, the traffic light detection system comprises a preset backbone network, a preset feature fusion module and a preset detection module, the preset backbone network is used for outputting preprocessed feature data and outputting first to-be-fused feature data according to original RAW domain data, outputting processed feature data according to the first to-be-fused feature data; the preset feature fusion module is used for outputting second to-be-fused feature data according to the processed feature data, outputting fused feature data according to the first to-be-fused feature data and the second to-be-fused feature data, and outputting to-be-detected feature data according to the fused feature data; and the preset detection module is used for outputting a category probability, a confidence coefficient and a target bounding box coordinate according to the to-be-detected feature data. By simplifying the number of network layers and reducing the number of convolution kernels, the detection precision of traffic light detection is improved while a simple and lightweight model is used.
Need to check novelty before this filing date? Find Prior Art