Surrounding Vehicle Motion Prediction in Non-Lane-Based Environments for ADAS Enhancement

The significance of forecasting the behavior of road agents is on the rise, particularly in Advanced Driver Assistance Systems (ADAS) enhancement. Predicting road agents’ intentions holds paramount importance for Autonomous vehicles, especially considering the forthcoming coexistence of ADAS systems with heterogeneous road entities within urban roadways. The behavioral attributes of non-lane based traffic, featuring a mix of various elements, are prevalent not only in urban scenarios but also in unstructured environments. This research aims to predict the movement patterns of surrounding vehicles in non-lane based environments. This study captures the surrounding vehicles’ intent to utilize tight lateral spaces in non-lane based environments through the variations of lateral descriptor values. The investigation takes into account several factors, including leveraging contextual cues, retaining spatial data related to neighboring vehicles, and identifying driving patterns. To achieve these objectives, a hybrid model is introduced, combining a modified structured Long Short Term Memory (LSTM) with lateral descriptor based uncertainty estimation on top of established detection and tracking algorithms. This integration enhances the ability to capture spatial attributes of neighboring vehicles along with the assessment of traffic conflict indicators with lateral descriptor contextual cues. The methodology is evaluated across two distinct datasets: one simulating scenarios of neighboring vehicles within well-defined urban road setups, and the other representing non-structured environments. The empirical findings highlight the effectiveness of the proposed method, showcasing an impressive 24.69% enhancement in prediction accuracy compared to baseline models with 5-seconds prediction horizon.