Urban traffic scene image understanding and multi-view crowd-sourcing optimization method

Abstract

The invention discloses an urban traffic scene image understanding and multi-view crowd intelligent optimization method, which utilizes intersection driving path prior modeling of road layout random variables and a road surface semantic prior modeling method based on an FCN learning network to improve the cognitive precision of an intersection driving path and a road surface and the semantic understanding degree of the whole scene. Cognitive-driven priori model and data-driven two-dimensional model Candidate box semantic features of three-dimensional space transformation are combined with a deep neural network to realize complex traffic scene three-dimensional target detection and recognition; The method comprises the following steps of: estimating and describing a three-dimensional sceneflow for a complex road environment such as an intersection traffic scene by cooperatively considering a vehicle position posture and driving track prior model, and comprehensively characterizing theposture and motion trend of a traffic participant in the scene; Aiming at the urban complex intersection road section environment, the holographic understanding of the traffic scene is realized through multi-view crowd-sourcing optimization, the traffic environment understanding is effective and easy to realize, and a decision basis is provided for the autonomous control of the intelligent vehicle.

Description

technical field [0001] The invention belongs to the technical field of traffic control, and in particular relates to image understanding of urban traffic scenes and multi-view group intelligence optimization methods. Background technique [0002] With the technological breakthroughs in 5G communication, big data, artificial intelligence and other fields, intelligent connected vehicles will become the commanding heights of my country's future strategy of seizing the automobile industry, and an important breakthrough for the transformation and upgrading of the national automobile industry from big to strong. It is of great strategic significance to shape the industrial ecology, promote national innovation, improve traffic safety, and achieve energy conservation and emission reduction. Intelligent networked vehicles are a combination of unmanned driving technology and networked communication technology, involving the interdisciplinary integration of multiple disciplines, among w...

AI Technical Summary

Problems solved by technology

Traditional object recognition methods such as target detection and recognition algorithms based on template matching, target detection and recognition algorithms based on HOG features + SVM classifiers, etc. cannot have good scalability and robustness because they only use the underlying information of the image

With the development of neural network and the emergence of regional convolutional neural network, the target detection and recognition algorithm based on convolutional neural network has better robustness, higher accuracy, and faster detection rate. However, in autonomous In the context of driving systems, the 2D image detection method based on monocular vision cannot solve the problem of accurately estimating the distance from the self-driving car to potential obstacles. To solve these problems, it is urgent to develop an intelligent network-oriented Effective and easy-to-implement traffic scene understanding method for automobiles, providing decision-making basis for autonomous control of intelligent vehicles

Images