Shayanta
Shopnil

Pedestrian trajectories are crucial for self-driving cars to plan their paths effectively. The sensors implanted in these self-driving vehicles, despite being state-of-the-art ones, often face inaccuracies in the perception of surrounding environments due to technical challenges in adverse weather conditions, interference from other vehicles’ sensors and electronic devices, and signal reception failure, leading to incompleteness in the trajectory data. But for real-time decision making for autonomous driving, trajectory imputation is no less crucial. Previous attempts to address this issue, such as statistical inference and machine learning approaches, have shown promise. Yet, the landscape of deep learning is rapidly evolving, with new and more robust models emerging. In this research, we have proposed an encoder–decoder architecture, the Human Trajectory Imputation Model, coined HTIM, to tackle these challenges. This architecture aims to fill in the missing parts of pedestrian trajectories. The model is evaluated using the Intersection drone the inD dataset, containing trajectory data at suitable altitudes, preserving naturalistic pedestrian behavior with varied dataset sizes. To assess the effectiveness of our model, we utilize L1, MSE, and quantile and ADE loss. Our experiments demonstrate that HTIM outperforms the majority of the state-of-the-art methods in this field, thus indicating its superior performance in imputing pedestrian trajectories.

Pedestrian trajectories are crucial for self-driving cars to plan their paths effectively. The sensors implanted in these self-driving vehicles, despite being state-of-the-art ones, often face inaccuracies in the perception of surrounding environments due to technical challenges for adverse weather conditions, interference from other vehicles sensors and electronic devices, and signal reception failure, leading to incompleteness in trajectory data. But for real-time decision making for autonomous driving, the trajectory imputation is no less crucial. Previous attempts to address this issue such as statistical inference and machine learning approaches have shown promise. Yet, the landscape of deep learning is rapidly evolving with new and more robust models emerging. In this research, we have proposed an encoder-decoder architecture, Human Trajectory Imputation Model, coined as HTIM, to tackle these challenges. This architecture aims to fill in the missing parts of pedestrian trajectories. The model has been evaluated using the Intersection drone inD dataset, containing trajectory data at suitable altitudes preserving naturalistic pedestrian behavior with varied dataset sizes. To assess the effectiveness of our model, we have utilized L1, MSE, Quantile and ADE loss. Our experiments have demonstrated that HTIM outperforms the majority of the state-of-the-art methods in this field, thus indicating its superior performance in imputing pedestrian trajectories.

Human Trajectory Imputation has been used for decades in surveillance, sports analytics, delivery and logistics, traffic management, etc. Recently, autonomous vehicles(AV) have entered into this realm. These vehicles have come to the transportation scenario for their safety, efficiency and convenience. These vehicles, filled with advanced sensors; analyze and navigate critical environments. GPS-based trajectory data, among these sensors, plays a vital role in motion planning. However, challenges such as signal loss due to obstructions, electromagnetic noise and technical constraints create in missing or erroneous trajectory points. Drones, positioned higher in the sky, offer advantages in signal reception and obstacle avoidance. That is why we are using inD dataset. The techniques addressed here tend to ensure reliable, real-time and accurate autonomous vehicle operation. With the stringent real-time demands …

In supervised learning, data classification is the method of categorizing data to facilitate data mining processes for informed decision-making. The central aim of a classification model is to accurately predict the categorical data for both familiar and unfamiliar instances. The classification models in machine learning are usually trained with datasets where instances are labeled. This paper explores an alternative way of constructing classification models based on the similarities of the instances rather than labels annotated by experts. The process of labeling data is a resource-intensive and time consuming process incredibly challenging when dealing with large datasets known as big data. In light of the proposed methodology clusters, the big data developing classifiers based on these clusters while bypassing the predefined class labels. This approach enhanced the performance of the classifier. Moreover, the …