Method and device for outputting information

Abstract

Embodiments of the present disclosure disclose a method and device for outputting information, which can be used in the field of automatic driving. A specific implementation of the method includes: in response to receiving the current speed information of the obstacle observed by the sensor, scoring the quality of the current speed information to obtain the current quality score; The speed information and quality score of the obstacle; if there is speed information in the cache queue that meets the predetermined condition and the quality score is higher than the current quality score, then use the speed information to replace the current speed information; perform filtering according to the current speed information, Obtain the fusion speed information of the obstacle and output it. This embodiment can realize timing verification of a single sensor and cross verification between different sensors, so as to achieve robustness to noise and complementary advantages between sensors.

Description

technical field [0001] Embodiments of the present disclosure relate to the field of computer technology, and in particular to methods and devices for outputting information. Background technique [0002] In the perception module of autonomous driving, multi-sensor fusion is often used as the last module of perception, receiving the results from various sensors and performing fusion output on these results, which plays a role of complementary advantages between different sensors. The speed attribute of the obstacle directly affects the downstream decision-making, and is one of the important attributes of the obstacle. Only by making full use of the speed estimation advantages of different sensors in different scenarios can the accurate and stable speed attribute estimation be output for the downstream. [0003] Existing multi-sensor speed fusion is usually carried out under a single filtering framework, and the speed fusion of sensors from different sources is realized by mo...

AI Technical Summary

Problems solved by technology

[0004] In the filtering framework, since the input sources of multi-sensor fusion come from different sensors, the heterogeneous timing input means that the Markov assumption is not true, that is, there is no dependence between the state and the state of the previous frame, and the motion model is difficult to be accurate. Characterize the timing relationship

At the same time, heterogeneous observations also bring difficulties to the modeling of observation models. The dimensions of observation noise given by different sensors are not consistent, and different sensors have different motion expression capabilities in different scenarios. The scene is expressed by observing noise changes. The difference of , appears unnatural and difficult to explicitly model numerically

In addition, in the case of unstable data association quality, the filtering framework can easily introduce noise, and it is difficult to capture abnormal observations due to unstable data association quality using the filtering framework

[0005] In terms of sensor capability complementarity, although the hybrid model framework can model the motion output of different sensors, such modeling does not make full use of the complementary advantages between different sensors, and still treats the output of each sensor separately. Instead of viewing multi-sensor observations as the movement expression of the same object at different angles

[0006] To sum up, the current technology is difficult to carry out scene-based fine modeling of the motion expression ability of each sensor, and the tolerance to noise is insufficient, and it is difficult to make full use of the capabilities of different sensors to complement each other.

Images