Zhang, Y., Zhang, S., & Luo, R. (2022). Lane Change Intent Prediction Based on Multi-Channel CNN Considering Vehicle Time-Series Trajectory. 2022 IEEE 25th International Conference on Intelligent Transportation Systems (ITSC). https://doi.org/10.1109/itsc55140.2022.9921941
Abstract:
Recent years, autonomous driving has always been a hot topic, which is regarded as an important means to shape the future of transport. However, safety is always the essential of any auto-pilot systems in the real-world. The safety and security of autonomous vehicles faces big challenges in public acceptance and trust. Thus, being able to accurately perceive the surrounding environment and respond in real time is an important goal of autonomous driving at the decision-making level. In this paper, a lane change prediction model using convolutional neural network is proposed combining the features from space and time. The proposed CNN learns the behavior pattern from historical data and capable of making lane change predictions to assist the decision making of auto-pilot systems in real-time. The experiments are conducted on the Next Generation Simulation (NGSIM) dataset, and we are able to predict up to 7s horizon before the actual lane-change took place with the observations of one second. An accuracy of 99 % is achieved for a 3 second observation window without further feature engineering. Comparisons are performed with tree based and multi-layer perceptron based methods.
License type:
Publisher Copyright
Funding Info:
There was no specific funding for the research done