Binocular semantic SLAM method for automatic driving scene

Abstract

The invention discloses a binocular semantic SLAM method for an automatic driving scene, and the method can detect and track other vehicles, pedestrians and other objects in a scene while enabling a vehicle to accurately track the pose of the vehicle in the environment in an automatic driving scene. According to the method, the quadric surface is used as the expression of a three-dimensional object, and compared with the prior art that the quadric surface can only be used for mapping or for a road sign for a static object, the quadric surface can be used for expressing the states of objects atdifferent moments no matter for the static object in a scene or a moving object. For the situation that semantic object observation is insufficient, the neural network and priori information are fully utilized for constraint, an error term for object observation is constructed, and nonlinear optimization solving of object state parameters is achieved.

Description

Technical field[0001]The invention relates to the field of simultaneous localization and map construction (SLAM), and in particular to a binocular semantic SLAM method for automatic driving scenarios.Background technique[0002]Simultaneous Localization and Mapping (SLAM) is one of the hot research topics in recent years. It refers to the use of equipment equipped with specific sensors to process the information received by the sensors in an unknown environment to estimate their own The location and orientation in the environment, as well as the algorithm to build a map of the environment in a specific format. As mobile robots, drones and other equipment increasingly appear in industrial and household environments, the field of augmented reality is booming, and more emerging technologies such as autonomous driving are in the ascendant, SLAM technology is one of the key algorithms in these fields. First, it has received more and more attention from industry and academia. Commonly used ...

AI Technical Summary

Problems solved by technology

The disadvantage of this method is that the orientation of the vehicle can only take discrete values, and the parameter expression is not precise enough; the mask generated by the three-dimensional bounding box is not fine enough, which is more likely to cause object mismatch; all objects are regarded as dynamic objects, and when the camera is tracking Unable to utilize information from static objects

Images