Pedestrian behavior intention prediction method based on multi-task learning

Abstract

The invention provides a pedestrian behavior intention prediction method based on multi-task learning. The method comprises the following steps: a training sample set is constructed; a pedestrian behavior intention prediction model is constructed by using the basic network, the attitude detection network and the intention recognition network, the basic network takes a single frame image in the training sample set as input, image features are extracted, and a feature map is obtained; an encoder part of the attitude detection network comprises a part intensity field sub-network and a part association field sub-network which respectively take the feature map as input and take the joint feature map and the skeleton feature map as output, and a decoder part of the attitude detection network obtains a pedestrian attitude image according to the joint feature map and the skeleton feature map; the intention recognition network takes the feature map as input and takes the pedestrian behavior intention image as output; and the pedestrian behavior intention prediction model is trained, and the pedestrian behavior intention is predicted by using the pedestrian behavior intention prediction model. According to the method, a pedestrian detector does not need to be adopted, each pixel is processed separately, the running time is constant, and the prediction effect meets the requirement.

Description

technical field [0001] The invention belongs to the technical field of autonomous decision-making of unmanned vehicles, in particular to a method for predicting pedestrian behavior intentions based on multi-task learning. Background technique [0002] The task of understanding and predicting pedestrian behavior intentions aims to identify pedestrian activity types and predict their action intentions before they make certain actions. This research content has a wide range of applications in the field of autonomous driving. With the frequent occurrence of traffic accidents, artificial intelligence technology urgently needs to be applied in the field of automatic driving. By predicting the behavioral intentions of pedestrians in advance, especially the intention of crossing the road, early decision-making can be made to avoid traffic accidents. However, due to the complexity of pedestrian behavior, and the high accuracy and efficiency requirements of the pedestrian behavior pr...

AI Technical Summary

Problems solved by technology

[0004] However, the above methods all require the use of pedestrian detectors. The accuracy of behavioral intention prediction is directly affected by the detection effect of pedestrian detectors, and the detection time will increase with the increase of the number of pedestrians in the scene, which will easily reduce the efficiency of model operation.

Images