BiLSTM-based weekly vehicle lane change intention prediction method

Abstract

The invention discloses a BiLSTM-based weekly vehicle lane change intention prediction method, and relates to the technical field of intelligent vehicles. The method comprises the following steps: 1) an offline training process of a lane change intention prediction model: constructing a training database through collection and processing of a large-scale real driving scene; and training a lane changing intention prediction model; and 2) online real-time prediction of the lane change intention prediction model: after real-time data acquisition and data preprocessing of the own vehicle, performing online real-time prediction by using the trained lane change intention prediction model to predict the current lane change intention of the surrounding vehicles. Factors influencing lane changing of surrounding vehicles are comprehensively considered, the Inception-ResNet-v2 network is utilized to extract frame level features of vehicle videos, the vehicle state and the frame level features are fused, the BiLSTM network prediction capability is improved, the detection precision and efficiency are further improved, the lane changing intention of the surrounding vehicles is effectively predicted, and a great role is played in improving the safety performance of the unmanned vehicle.

Description

technical field [0001] The invention relates to the technical field of intelligent vehicles, in particular to a method for predicting lane-changing intentions based on BiLSTM. Background technique [0002] In the future, unmanned vehicles will operate in a human-machine mixed traffic environment for a long time, that is, a traffic environment in which human-driven vehicles and unmanned vehicles are mixed in the road system. However, due to the complexity of the actual road traffic environment and the variability of the driver's driving style, the interaction between unmanned vehicles and surrounding vehicles has become a current research difficulty. According to the investigation of the traffic control department, lane-changing accidents account for about 30% of the total number of all types of traffic accidents. Therefore, studying the prediction of vehicle lane-changing intention is of great significance for road safety and improving the ability of unmanned vehicle decisi...

AI Technical Summary

Problems solved by technology

Literature [1] (Lu C, etc. Virtual-to-Real Knowledge Transfer for Driving Behavior Recognition: Framework and a Case Study [J]. IEEE Transactions on Vehicular Technology, 2019, 68(7): 6391-6402.) proposed a A lane-changing intention recognition method based on transfer learning, the interaction information only considers the longitudinal relative positional relationship between the ego vehicle and the vehicle in front, literature [2] (Ding W, etc. Predicting Vehicle Behaviors Over An Extended Horizon Using Behavior InteractionNetwork, 2019International Conference on Robotics and Automation (ICRA). 2019, Montreal, Canada.) proposed a method for predicting lane-changing intentions based on recurrent neural networks. Interactive information, but only rely on the historical state information of the vehicle to predict the lane change intention, the target feature detection is simple and the practicability is not good

Images