Effects of CNN Backbone on Trajectory Prediction Models for Autonomous Vehicle

Seoyoung Lee,Hyogyeong Park,Yeonhwi You,Sungjung Yong,Il-Young Moon 한국정보통신학회JICCE 2023 Journal of information and communication convergen Vol.21 No.4

Trajectory prediction is an essential element for driving autonomous vehicles, and various trajectory prediction models have emerged with the development of deep learning technology. Convolutional neural network (CNN) is the most commonly used neural network architecture for extracting the features of visual images, and the latest models exhibit high performances. This study was conducted to identify an efficient CNN backbone model among the components of deep learning models for trajectory prediction. We changed the existing CNN backbone network of multiple-trajectory prediction models used as feature extractors to various state-of-the-art CNN models. The experiment was conducted using nuScenes, which is a dataset used for the development of autonomous vehicles. The results of each model were compared using frequently used evaluation metrics for trajectory prediction. Analyzing the impact of the backbone can improve the performance of the trajectory prediction task. Investigating the influence of the backbone on multiple deep learning models can be a future challenge.

Vehicle trajectory prediction based on Hidden Markov Model

( Ning Ye ),( Yingya Zhang ),( Ruchuan Wang ),( Reza Malekian ) 한국인터넷정보학회 2016 KSII Transactions on Internet and Information Syst Vol.10 No.7

In Intelligent Transportation Systems (ITS), logistics distribution and mobile e-commerce, the real-time, accurate and reliable vehicle trajectory prediction has significant application value. Vehicle trajectory prediction can not only provide accurate location-based services, but also can monitor and predict traffic situation in advance, and then further recommend the optimal route for users. In this paper, firstly, we mine the double layers of hidden states of vehicle historical trajectories, and then determine the parameters of HMM (hidden Markov model) by historical data. Secondly, we adopt Viterbi algorithm to seek the double layers hidden states sequences corresponding to the just driven trajectory. Finally, we propose a new algorithm (DHMTP) for vehicle trajectory prediction based on the hidden Markov model of double layers hidden states, and predict the nearest neighbor unit of location information of the next k stages. The experimental results demonstrate that the prediction accuracy of the proposed algorithm is increased by 18.3% compared with TPMO algorithm and increased by 23.1% compared with Naive algorithm in aspect of predicting the next k phases` trajectories, especially when traffic flow is greater, such as this time from weekday morning to evening. Moreover, the time performance of DHMTP algorithm is also clearly improved compared with TPMO algorithm.

Multi-modal Pedestrian Trajectory Prediction based on Pedestrian Intention for Intelligent Vehicle

Youguo He,Yizhi Sun,Yingfeng Cai,Chaochun Yuan,Jie Shen,Liwei Tian 한국인터넷정보학회 2024 KSII Transactions on Internet and Information Syst Vol.18 No.6

The prediction of pedestrian trajectory is conducive to reducing traffic accidents and protecting pedestrian safety, which is crucial to the task of intelligent driving. The existing methods mainly use the past pedestrian trajectory to predict the future deterministic pedestrian trajectory, ignoring pedestrian intention and trajectory diversity. This paper proposes a multimodal trajectory prediction model that introduces pedestrian intention. Unlike previous work, our model makes multi-modal goal-conditioned trajectory pedestrian prediction based on the past pedestrian trajectory and pedestrian intention. At the same time, we propose a novel Gate Recurrent Unit (GRU) to process intention information dynamically. Compared with traditional GRU, our GRU adds an intention unit and an intention gate, in which the intention unit is used to dynamically process pedestrian intention, and the intention gate is used to control the intensity of intention information. The experimental results on two first-person traffic datasets (JAAD and PIE) show that our model is superior to the most advanced methods (Improved by 30.4% on MSE0.5s and 9.8% on MSE1.5s for the PIE dataset; Improved by 15.8% on MSE0.5s and 13.5% on MSE1.5s for the JAAD dataset). Our multi-modal trajectory prediction model combines pedestrian intention that varies at each prediction time step and can more comprehensively consider the diversity of pedestrian trajectories. Our method, validated through experiments, proves to be highly effective in pedestrian trajectory prediction tasks, contributing to improving traffic safety and the reliability of intelligent driving systems.

지역 환경 및 차량 상호작용을 활용한 two-stage 멀티모달 차량 미래 경로 예측 기술

최세환(Sehwan Choi),윤준용(Junyong Yun),김정호(Jungho Kim),최준원(Jun Won Choi) 한국자동차공학회 2023 한국자동차공학회 부문종합 학술대회 Vol.2023 No.5

In this paper, we propose a two-stage multi-modal future trajectory prediction framework designed to effectively utilize significant inter-agent interaction and local scene context. This two-stage motion prediction architecture, referred to as GL-Pred, consists of two networks: the proposal trajectory network and the refinement trajectory network. The proposal trajectory network produces multi-modal trajectory proposals by leveraging past trajectories and global environmental information. The refinement trajectory network enhances each of the trajectory proposals using group-query attention and localquery attention mechanisms. Group-query attention mechanism further enhances the trajectory proposals by modeling the interagent interaction by grouping the proposal trajectory of the neighboring agents. Local-query attention mechanism is used to aggregate local scene context features collected from around trajectory proposals. Finally, we combine group-query attention and local-query attention features to produce the multi-modal future trajectory. The experiments conducted on the Argoverse dataset demonstrate that the proposed GL-Pred outperforms existing motion prediction methods.

Trajectory Pattern Construction and Next Location Prediction of Individual Human Mobility with Deep Learning Models

Dabin You,Ha Yoon Song 한국정보과학회 2020 Journal of Computing Science and Engineering Vol.14 No.2

Many modern portable devices, especially smartphones, are equipped with positioning functionality. The rapid growth in the use of such devices has allowed for the accumulation of a vast amount of positioning data. Combined with deep learning methods, these data may be used for many novel applications. Herein, a trajectory pattern tree generation method via deep learning is proposed. The convolutional neural network (CNN) and recurrent neural network (RNN) model of deep learning were applied for trajectory generation and prediction. Several volunteers provided their raw positioning data. The trajectory generation and prediction are for individual mobility patterns and were performed for every volunteer. We present the results obtained from seven volunteers. The preciseness of prediction can be measured both for CNN and RNN. Consequently, we can predict an individual’s location with 32.98% accuracy, and predict the top-five up to 69.22% for unit area size of 0.030 km².

Trajectory-prediction based relay scheme for time-sensitive data communication in VANETs

( Zilong Jin ),( Yuxin Xu ),( Xiaorui Zhang ),( Jin Wang ),( Lejun Zhang ) 한국인터넷정보학회 2020 KSII Transactions on Internet and Information Syst Vol.14 No.8

In the Vehicular Ad-hoc Network (VANET), the data transmission of time-sensitive applications requires low latency, such as accident warnings, driving guidance, etc. However, frequent changes of topology in VANET will result in data transmission failures. In order to improve the efficiency of VANETs data transmission and increase the timeliness of data, this paper proposes a relay scheme based on Recurrent Neural Network (RNN) trajectory prediction, which can be used to select the optimal relay vehicle to transmit data. The proposed scheme learns vehicle trajectory in a distributed manner and calculates the predicted trajectory, and then the optimal vehicle can be selected to complete the data transmission, which ensures the timeliness of the data. Finally, we carry out a set of simulations to demonstrate the performance of the algorithm. Simulation results show that the proposed scheme enhances the timeliness of the data and the accuracy of the predicted driving trajectory.

Modeling User Trajectory Similarity for Next Location Prediction

Yixuan Ge,Pengcheng Li 한국차세대컴퓨팅학회 2022 한국차세대컴퓨팅학회 학술대회 Vol.2022 No.10

With the rapid development of social media, users' next location prediction has become an important research direction, which can provide personalized travel suggestions for users. However, existing methods ignore the semantic relationship between users' historical and current trajectories. This paper proposes a new method for predicting the user's next location to solve this problem. We first process the user POI data as trajectory data, use the attention mechanism to extract similar features of the user's historical trajectories, and then combine them with the current trajectory features to obtain the user's next location recommendation. The experimental results show that our proposed model performs satisfactorily on a real dataset.

도착관리시스템을 위한 항적 자료 기반 궤적 예측 연구

오은미,전대근,김현경,은연주 한국항공우주학회 2014 한국항공우주학회 학술발표회 논문집 Vol.2014 No.11

도착관리시스템의 관제 지원 조언 정보를 생성하기 위해 항적 자료에 기반한 궤적 예측 방법에 대한 연구를 제안하였다. 제안된 궤적 예측 소프트웨어는 궤적을 계산하여 주요 픽스에 대한 예상 도착 시간정보를 계산하며 스케줄링에 필요한 정보를 함께 제공한다. 항공기의 궤적은 AFTN 및 비행계획 처리기에서 제공된 비행 계획에 의거하여 비행하는 것을 전제로, 감시 자료 처리기로부터 실시간으로 전달되는 항적 자료와 비교를 통해 궤적 갱신 여부를 결정하고 궤적을 재계산할 경우 항적 자료의 항공기 상태 정보를 적용하여 궤적 예측의 정확도를 높이고자 하였다. 시뮬레이션을 통해 제안된 방법에 대한 실행 결과를 확인 하였으며 이는 차후 실제 데이터의 궤적 예측 연구에 활용될 예정이다. A study of trajectory prediction which provides advisory information to support arrival control is introduced. The trajectory predictor provides ETA and other time-based information for scheduling. The trajectory is calculated using the track data from surveillance data processor and its flight plan given by AFTN(Aeronautical Fixed Telecommunication Network) and FDP(Flight Data Processor). It is assumed that the aircraft complies with flight plan and every time the track data is inputted, the trajectory predictor compares the data with the predicted trajectory. The simulation shows suggested method of trajectory prediction could be implemented for AMAN. For the next step, it is planned to be tested with real track data.

궤적 기반의 항공 교통 관리를 위한 스케줄링 시스템 개발

오은미 ( Eun-mi Oh ),은연주 ( Yeonju Eun ),김현경 ( Hyounkyoung Kim ),전대근 ( Daekeun Jeon ) 한국항행학회 2018 韓國航行學會論文誌 Vol.22 No.5

차세대 통신링크를 활용한 항공로 교통 관리를 위해 궤적 기반의 관제 지원 스케줄링 시스템을 제안하였다. 차세대 ATS (air traffic services) 데이터링크인 Baseline 2를 사용하는 4DTRAD (4-dimensional trajectory data link) 서비스 내용을 기반으로 항공로상을 비행 중인 항공기를 대상으로 하는 궤적기반운용 수행 절차를 수립하고 기술하였다. 이러한 절차를 바탕으로, 다양하고 복잡한 데이터 활용으로 인한 관제사의 업무 부담을 완화하기 위해 지상 시스템이 수신한 항공기 데이터를 처리하여 궤적을 예측하고 관제 조언 정보를 제공하는 스케줄링 시스템의 프로토타입을 개발하였다. 또한, 궤적 기반 항행을 위한 시뮬레이션 환경을 구성하여 개발 시스템에 대한 스케줄링 기능을 확인하였다. A trajectory-based scheduling system is proposed for air traffic management using next generation aviation data communication link. Based on the service concept of 4-dimensional trajectory data link (4DTRAD) using air traffic serveices (ATS) datalink Baseline 2, a procedure for trajectory-based operation of an en-route flight is established and described in detail. To mitigate air traffic controllers’ workload which might be caused by various and complicated data utilization, a prototype of the scheduling system, which predicts the aircraft trajectory based on the flight intents received by air traffic service system and provides advisory information for air traffic control, was developed. The simulation environment for trajectory based operation was built to validate the scheduling functionality of the prototype.

PREDICTIVE CONTROL OF A VEHICLE TRAJECTORY USING A COUPLED VECTOR WITH VEHICLE VELOCITY AND SIDESLIP ANGLE

이정한,유완석 한국자동차공학회 2009 International journal of automotive technology Vol.10 No.2

In this paper, a predictive algorithm for vehicle trajectory control using the vehicle velocity and sideslip angle is proposed. Since the driving state of a vehicle generates nonholonomic constraint equations, it is difficult to control the trajectory with a conventional control algorithm. Furthermore, control vectors such as vehicle velocity and sideslip angle are coupled together; hence, a separate control for each variable is not suitable. In this study, a coupled control vector that combines the velocity and sideslip angle is proposed for the predictive control of vehicle trajectory. Since the coupled control vector is derived from the status of the vehicle’s motion, it is easy to generate a feedback control vector for the predictive controller. The coupled vector cannot be directly used as input to the vehicle systems; therefore, the vehicle input vector should be calculated from the control vector using a nonlinear function. Since nonlinear functions are not inserted in the control loop, they are calculated by the controller. Therefore, this method does not require a linearization process in the control logic, which enhances the stability and accuracy of the predictive controller.