Phase 1 Draft by Yashu (log)

[wangyashuu]

Phase 1 Report (Draft Version)

Goal

Phase I) Problem and Workflow Definition (Due on 21.11.2021)

• Give an introduction and motivation for your problem
• Give a problem description and definition
• Formulate a research goal and research questions
• Start a first related work search and provide your team with initial resources

Draft

Introduction

• Trajectory Prediction is useful: application for Service robots, Self-driving vehicles, Surveillance (ref: Human Motion Trajectory Prediction: A Survey)
• Trajectory Prediction is useful in the safety of “intelligent” vehicles) (ref: A survey on motion prediction and risk assessment for intelligent vehicles)
• Recent research shows GAN improve performence for Trajectory Prediction (ref: Goal-GAN, MG-GAN, Social GAN)

Motivation

• inference of the intensions behide trajectories can improve the performance of some application like security surveillance and auto vehicle safety systems to produce an earlier warning
• Try to get the intensions might helps at the predictions of trajectories, since human trajectories is complicated and has many hidden reasons.

Proble description and definition

• get an acceptable prediction of trajectories (we are not aim to have a high accuracy of prediction)
• derivate intensions of the trajectories

Related Work

• Sequence Model (it is a typical sequence model)

  • Social LSTM
  • LSTM An LSTM Network for Highway Trajectory Prediction

• GAN (reseach shows good at Trajectory Prediction)

  • SoPhie (lstm + attention + GAN)
  • Social GAN

• InfoGAN (infer the intension)

  • Social Ways

TODO

• What affects trajectory?

Links

• Trajectory Prediction on ETH/UCY

[wangyashuu]

for log