Research On Pedestrian Crossing Intent Prediction And Track Prediction Method In Mixed Traffic Scene

Autonomous driving technology has great development potential in enhancing driving safety and improving road traffic efficiency.It is attracting the attention of scholars at home and abroad,and has gradually become the focus of domestic and foreign automobile industry research.Pedestrian behavior prediction and trajectory prediction tasks,as high-level environmental perception technologies,rely on the results of primary perception algorithms such as detection and segmentation as input,and the output prediction results provide necessary input for subsequent planning decisions and motion control algorithms.As the key technology connecting the preceding and the following in the driving technology system,improving the prediction accuracy of pedestrian crossing intention is of great significance to driving safety.However,pedestrian movement has a high degree of uncertainty,and the complex road traffic environment in the real scene also brings great challenges to the perception task.Most of the existing methods are aimed at the bird’s-eye view,and rely too much on the pedestrian’s bounding box position trajectory information,there are few algorithms that focus on the solution from the perspective of the vehicle camera;at the same time,most of the existing methods have the problem of insufficient utilization of the interaction relationship in the road traffic scene;Interrelationships between trajectories.Aiming at the above problems,this paper proposes a multi-task method for pedestrian crossing intention and future trajectory prediction based on multimodal feature fusion from the perspective of vehicle-mounted cameras,which realizes accurate pedestrian crossing intention and future trajectory prediction in complex traffic scenes.predict.This paper is mainly supported by the sub-project of the National Key R&D Program "Research on Multi-target Behavior Recognition and Prediction Algorithms around Vehicles".The main research work is as follows:(1)First of all,a new global scene context interaction information extraction module is established,which includes dilated convolution,SE module and interactive attention mechanism,and relies on the scene semantic mask to model the interaction relationship between pedestrians and traffic elements;at the same time,A local scene spatiotemporal feature extraction module is constructed,which combines channel attention,spatial attention and AUGRU,and based on the multiple attention mechanism,improves the ability of the model to capture spatiotemporal information of local traffic scenes in complex traffic scenes.The proposed model innovatively combines a variety of attention mechanisms to enhance its ability to extract spatio-temporal features,and relies on the semantic analysis results of the scene to capture the interaction between pedestrians and their surroundings,solving the problem of traffic environment context information and traffic objects.The problem of insufficient application of interactive information between.(2)A multi-modal feature fusion module based on a hybrid fusion strategy is designed,which realizes the joint reasoning of visual features and motion features according to the complexity of different information sources,and provides reliable information for the pedestrian crossing intention prediction module.The test based on the JAAD dataset shows that the prediction Accuracy of the proposed method is 0.84,which is 10.5% higher than the baseline method.Compared with the existing models of the same type,the proposed method has the best overall performance and has broader application scenario.(3)On the basis of the pedestrian crossing intention prediction algorithm proposed in this paper,the concept of multi-task network is introduced,and the multi-task network of layered fusion is constructed by fully utilizing the internal linkage relationship between pedestrian behavior intention and pedestrian movement trajectory through feature coupling.,the model can realize the prediction of the pedestrian’s future bounding box position trajectory while predicting the crossing intention of the target pedestrian in front of the vehicle in complex traffic scenarios.