University of Illinois Urbana-Champaign

Detecting the unpredictable: Advanced trajectory and camera-based anomaly detection methods for vehicles

Chakraborty, Neeloy

Content Files
CHAKRABORTY-THESIS-2023.pdf

Permalink

https://hdl.handle.net/2142/120160

Description

Title
Detecting the unpredictable: Advanced trajectory and camera-based anomaly detection methods for vehicles
Author(s)
Chakraborty, Neeloy
Issue Date
2023-05-02
Keyword(s)
  • Anomaly Detection
  • Autonomous Vehicles
  • Unsupervised Learning
  • Human Behavior Modeling
  • Deep Computer Vision
Abstract
In autonomous driving, detection of abnormal driving behaviors is essential to ensure the safety of vehicle controllers. Prior works in vehicle anomaly detection have shown that modeling interactions between agents improves detection accuracy, but certain abnormal behaviors where structured road information is paramount are poorly identified, such as wrong-way and off-road driving. We first propose a novel unsupervised framework for trajectory-based highway anomaly detection named Structural Attention-Based Recurrent VAE (SABeR-VAE), which explicitly uses the structure of the environment to aid anomaly identification. Specifically, we use a vehicle self-attention module to learn the relations among vehicles on a road, and a separate lane-vehicle attention module to model the importance of permissible lanes to aid in trajectory prediction. Conditioned on the attention modules' outputs, a recurrent encoder-decoder architecture with a stochastic Koopman operator-propagated latent space predicts the next states of vehicles. Our model is trained end-to-end to minimize prediction loss on normal vehicle behaviors, and is deployed to detect anomalies in (ab)normal scenarios. While SABeR-VAE outperforms state-of-the-art approaches, a pure trajectory sensor source is infeasible to access in real-world deployment. Thus, we then explore methods for camera-based anomaly detection conditioned on image sources. We apply convolutional autoencoder approaches and a Koopman model to the Detection of Traffic Anomaly (DoTA) dataset to evaluate methods on real anomalous scenarios captured on video.
Type of Resource
Thesis
Handle URL
https://hdl.handle.net/2142/120160
Copyright and License Information
Copyright 2023 Neeloy Chakraborty

Owning Collections

Proceedings of the ALISE Annual Conference: ALISE 2022 PRIMARY
Go Back and Get It: From One Narrative to Many