Review of Pedestrian Trajectory Prediction Methods: Comparing Deep Learning and Knowledge-based Approaches

Abstract

In crowd scenarios, predicting trajectories of pedestrians is a complex and challenging task depending on many external factors. The topology of the scene and the interactions between the pedestrians are just some of them. Due to advancements in data-science and data collection technologies deep learning methods have recently become a research hotspot in numerous domains. Therefore, it is not surprising that more and more researchers apply these methods to predict trajectories of pedestrians. This paper compares these relatively new deep learning algorithms with classical knowledge-based models that are widely used to simulate pedestrian dynamics. It provides a comprehensive literature review of both approaches, explores technical and application oriented differences, and addresses open questions as well as future development directions. Our investigations point out that the pertinence of knowledge-based models to predict local trajectories is nowadays questionable because of the high accuracy of the deep learning algorithms. Nevertheless, the ability of deep-learning algorithms for large-scale simulation and the description of collective dynamics remains to be demonstrated. Furthermore, the comparison shows that the combination of both approaches (the hybrid approach) seems to be promising to overcome disadvantages like the missing explainability of the deep learning approach.

Mapping the knowledge domain of soft computing applications for emergency evacuation studies: A scientometric analysis and critical review

“…The term "machine" here represents an intelligent system that makes decisions autonomously and independently (wholly or partly), and its autonomy is primarily achieved through simulation, artificial intelligence, and heuristic algorithms (Roth et al, 2019;Xiong et al, 2022;Li et al, 2022a). In order to make an evacuation decision, expert experience plays a critical role, especially in the preparation of evacuation plans, safety evacuation regulations, and rescue resource allocation, and often relies on expert knowledge and previous experience (Korbmacher and Tordeux, 2021). However, when facing large-scale complex evacuation problems, expert knowledge systems often fail to make optimal decisions.…”

Prediction of Social Dynamic Agents and Long-Tailed Learning Challenges: A Survey

“…Georgiou et al (2018) review trajectory prediction of pedestrians, cars, boats, planes, and animals on datasets primarily involving maps and GPS coordinates while Gulzar et al (2021) cover both physics and learning based modeling approaches on pedestrians and vehicles across various modalities (map aware, scene aware, interaction aware) and task types (intent prediction, trajectory prediction, occupancy maps). There are also vehicle specific (Leon & Gavrilescu, 2021;Paravarzar & Mohammad, 2020) and pedestrian specific (Korbmacher & Tordeux, 2021;Rudenko et al, 2020) surveys. The most notable out of these are and Rudenko et al (2020).…”

Pedestrian Behavior Prediction Using Deep Learning Methods for Urban Scenarios: A Review

“…Rudenko et al [16] reviewed the work related to human motion trajectory prediction and categorized existing methods by the modeling approach and contextual cues. Korbmacher and Tordeux [17] reviewed pedestrian trajectory prediction methods, compared deep learning methods and knowledge-based methods. These papers only covered the trajectory prediction and omitted the important prediction of intention that can be used for pedestrian-vehicle collision avoidance.…”

“…Ridel et al [18] reviewed and classified existing pedestrian behavior prediction models, but they classified previous works from only a single criterion, and many recently suggested deep learning methods were not covered. Most of the previous review papers focused on a single task, either the analysis of trajectories [15], [16], [17] or intention [9], or interactions between pedestrians and vehicles [10], [11], which did not cover the aspects in this paper's scope. Moreover, most of these papers classified the existing literature by a single criterion [17], [18], and only include methods with some particular input data [15].…”

“…Most of the previous review papers focused on a single task, either the analysis of trajectories [15], [16], [17] or intention [9], or interactions between pedestrians and vehicles [10], [11], which did not cover the aspects in this paper's scope. Moreover, most of these papers classified the existing literature by a single criterion [17], [18], and only include methods with some particular input data [15].…”