A general, context-aware pedestrian trajectory prediction model

Autonomous driving on highways and freeways, as a feature, is already available in quite a few high-end commercial vehicles being sold today. Autonomous driving in urban environments, on the other hand, is still an active area of academic and industrial research [6], because of its relatively complex nature. Urban driving requires the self-driving vehicle to interact with not just other vehicles, but also other moving agents such as cyclists and pedestrians. Pedestrian trajectory prediction is challenging because of the relatively higher number of degrees of freedom in pedestrian movement and absence of uniform rules across different cities and different scenarios within a city. Furthermore, in scenarios such as intersections, context, such as pedestrian traffic lights, stop signs and sidewalk geometry, significantly influences pedestrian movement. The objective of this thesis is to present a general, context-aware, long term (order of few seconds) trajectory prediction model for pedestrians in urban intersections. To meet this objective, first, the Augmented Semi Nonnegative Sparse Coding (ASNSC) [13] framework, for trajectory prediction, is extended to embed context, and build the Context-aware Augmented Semi Nonnegative Sparse Coding (CASNSC) algorithm. For prediction in new, unseen intersections with different curbside geometries (orthogonal versus skewed), CASNSC is further extended to build the Transferable Augmented Semi Nonnegative Sparse Coding (TASNSC) algorithm. Urban intersections can at times vary significantly in the type of pedestrian behaviors encountered, even across intersections with similar geometries. For instance, faster, rule breaking students near a college campus versus slower pedestrians in a residential area. While TASNSC is capable of successfully transferring knowledge from one intersection to another, it lacks the ability to update its prediction model as, and when, new intersections are visited and novel behaviors are encountered. An online model, based on TASNSC, is also presented in this thesis to account for this particular limitation.

Department dc:contributor.department

Massachusetts Institute of Technology. Department of Mechanical Engineering.

Grantor dc:publisher

Massachusetts Institute of Technology

Year dc:date.issued

2018

Author dc:creator

• Jaipuria, Nikita

Advisor dc:contributor.advisor

• Jonathan P. How.

• Mechanical Engineering.

Statement dc:rights

• MIT theses are protected by copyright. They may be viewed, downloaded, or printed from this source but further reproduction or distribution in any format is prohibited without written permission.

Licence dc:rights.uri

• http://dspace.mit.edu/handle/1721.1/7582

Language dc:language.iso

eng

Handle dc:identifier.uri

http://hdl.handle.net/1721.1/120227

OAI identifier oai:identifier

oai:dspace.mit.edu:1721.1/120227