A road environment visual perception method based on an improved Faster R-CNN

Abstract

The present invention relates to a road environment visual perception method based on an improved Faster R-CNN. Aiming at high-precision requirements of target detection and identification in a complex road scene, the invention provides an improved Faster R-CNN algorithm based on multi-GPU training. According to the algorithm, a multi-GPU parallel training method is used, so that the training efficiency is improved. By employing the ResNet-101 feature extraction network, the target detection precision is improved. By employing the Soft-NMS algorithm to reduce the omission factor, at the same time, introducing an OHEM into the ROI NetWork to reduce the false alarm rate, in order to improve the target detection effect of the algorithm in rainy, snowy and haze weather, the model is trained bycombining an internationally recognized automatic driving data set KITTI and Oxford RobotCar. The experimental results prove that by comparing with the Faster R-CNN, the algorithm has the advantagesthat the training speed and the detection precision are obviously improved, and particularly, and has good generalization ability and stronger practicability in an automatic driving scene.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to a road environment visual perception method based on improved Faster R-CNN. Through the improvement of the Faster R-CNN algorithm, this method has significantly improved the network model training speed and target detection accuracy, especially in automatic driving scenarios with good generalization ability and stronger practicability. Background technique [0002] The self-driving car milestone began in 2009 when Google began developing a self-driving car project known as Waymo. In recent years, as AlphaGo demonstrated the powerful learning ability of deep learning, the application of environment perception and driving decision-making algorithms based on deep learning in automatic driving has made it possible for unmanned driving to truly replace human driving. However, in the face of complex road scenes, the safety and reliability problems of autonomous dr...

AI Technical Summary

Problems solved by technology

However, in the face of complex road scenes, the safety and reliability problems of autonomous driving technology have always been the bottleneck restricting the promotion and popularization of driverless cars

However, the Faster R-CNN algorithm also has many shortcomings

For example, Faster R-CNN can only use a single GPU for training. When the number of training samples is huge or the feature extraction network is deepened, it will cause problems such as long training time and insufficient video memory; because the features extracted by the feature extraction network are not fine, resulting in The problem of target missed detection; when encountering complex scenes or targets with occlusion or deformation, the problem of target prediction errors

Images