Advanced driver assistance and automated driving systems should be capable of predicting and avoiding dangerous situations. In this study, we propose a method to predict potentially dangerous lane changes (cut-ins) of the vehicles in front. We follow a computer vision-based approach that only employs a single in-vehicle RGB camera and we classify the target vehicle's maneuver based on the recent video frames. Our algorithm consists of a CNN-based vehicle detection and tracking step and an LSTM-based maneuver classification step. It is computationally efficient compared to other vision-based methods since it exploits a small number of features for the classification step rather than feeding CNNs with RGB frames. We evaluated our approach on a publicly available driving dataset and a lane change detection dataset. We obtained 0.9585 accuracy with side-aware two-class (cut-in vs. lane-pass) classification models. Experiment results also reveal that, when used for lane change detection, our approach outperforms state-of-the-art approaches.
Proposed method architecture:
Proposed pipeline's classification code is available in "maneuver_prediction_train_and_test_LSTM.py".
After you extracted target vehicle bounding box features and created your train-validation-test X and y csv files as mentioned in the paper, you can get evaluation results from the code.
About the data we used:
- First, you should create a user at Berkeley Deep Drive Dataset portal (https://bdd-data.berkeley.edu/).
- After you download the training data from BDD-100K, you can cut cut-in and lane-pass maneuvers from them by using the data.csv file.
Note: information about data.csv file will be here after the conference.