Semantic recognition method, system and device for vehicle-mounted camera

Abstract

The invention relates to a semantic recognition method for a vehicle-mounted camera, and the semantic recognition method comprises the steps: collecting an original image through the vehicle-mounted camera, carrying out the distortion correction of the original image, and obtaining a video image facilitating the subsequent processing; and then, performing pixel-level semantic recognition on a target in the video image, meanwhile, utilizing a multi-task architecture of a network for training to obtain a model so as to output scene information of the video image, and then tracking the target based on the scene information. The semantic recognition method for a vehicle-mounted camera can adapt to various scenes and has the advantages of being high in precision, high in speed, wide in application range and the like. The invention further relates to a semantic recognition system for a vehicle-mounted camera and a semantic recognition device for a vehicle-mounted camera executing the semantic recognition method.

Description

technical field [0001] The present application relates to the field of vehicles, and in particular to a vehicle-mounted camera semantic recognition method for driving assistance, a vehicle-mounted camera semantic recognition system and a vehicle-mounted camera semantic recognition device for implementing the above method. Background technique [0002] With the development of computer computing power and technological progress, autonomous driving has also developed rapidly. Semantic segmentation and semantic recognition, as important modules in the field of autonomous driving, have also developed rapidly. Semantic segmentation is to judge the category to which each pixel in the image belongs, such as road surface, curb stone, person, car and other objects. At present, the commonly used sensors for semantic recognition mainly include stereo binocular cameras, lidar, and monocular cameras. These sensors have achieved good accuracy, but each has its own drawbacks. The binocul...

AI Technical Summary

Problems solved by technology

But this method is in practical application, owing to being subjected to the difference of surrounding environment, as underground garage, park road, underground garage, urban road etc., and the difference of weather, as sunny day, rainy day (especially can run into ground water in rainy day, causes specular reflection) etc.

In actual application, there are defects of poor robustness and low recognition accuracy

Images