Model training method and device and unmanned vehicle motion strategy determining method and device

Abstract

The invention discloses a method and a device for training a model and determining a motion strategy of an unmanned vehicle, and the method comprises the steps: determining a plurality of groups of continuous two-frame images in each continuous frame image collected in history as each training sample, determining a bounding box of a second image as a label for each training sample, and through a preprocessing model, determining a pre-processing result, determining a prediction result and a post-processing result of a first image of a training sample through a detection model, determining a detection result through a classification layer, and training a to-be-trained detection model by taking the minimum difference between the detection result and a label as an optimization target. According to the method, the obstacle instances of the first image in the training samples are predicted, at least part of the to-be-selected bounding boxes of the obstacle instances of the second image are updated, finally the bounding boxes of the obstacle instances of the second image are determined through the classification layer and serve as the detection result, manual parameter adjustment is not needed, and the accuracy is higher.

Description

technical field [0001] This description relates to the field of unmanned driving technology, and in particular to a method and device for model training and determination of unmanned vehicle motion strategy. Background technique [0002] At present, in order to ensure the driving safety of the unmanned vehicle, during the driving process of the unmanned vehicle, the obstacle instance around the unmanned vehicle is usually detected, and according to the position of the obstacle instance around the unmanned vehicle, the unmanned vehicle is determined. movement strategy to avoid obstacles. [0003] Generally, an unmanned vehicle can be equipped with a sensor for collecting images. When it is necessary to determine a movement strategy, it can first perform image detection on the images collected at each time in the most recent period, and correlate the detection results at each time to realize The detection and tracking of obstacle instances in each image, based on the detectio...

AI Technical Summary

Problems solved by technology

[0006] However, in the process of realizing the detection and tracking of obstacle instances in the prior art, the adjustment of the parameters of the Kalman filter algorithm and the selection of features in the similarity matching need to be set according to human experience, resulting in relatively low accuracy. Low

Images