Feasibility Study: Al-based Passenger Counting System for Public Transit
This thesis is dedicated to the subject of artificial intelligence and how it might contribute to public transport. Thereby, AI based object recognition for counting passengers in public transport shall be applied and compared with the existing infrared counting technology.
Peraic, Martin Hrvoje, 2019
Art der Arbeit Bachelor Thesis
Auftraggebende Basler Verkehrs-Betriebe
Betreuende Dozierende Jüngling, Stephan
Keywords AI, deep learning, computer vision, automated passenger
counting system, public transportation, feasibility study,
mobileSDD, Google Coral, Edge Device Computing, TNW,
Views: 48 - Downloads: 11
Over 350 000 passengers are transported daily by the Basler Verkehrs-Betriebe.
Many of these passengers hold a season ticket, which is offered by the Tarifverbund
Nordwerstschweiz (TNW). The TNW compensates all public transport providers
based on the number of passengers transported. Therefore, all operators must
provide verifiable passenger counts. Recent breakthroughs in AI based object
recognition enable reliable real-time object detection. Combined with a tracking, the
aim is to determine whether this technology can be used to count passengers in
public transport.
Various edge devices optimized for AI were evaluated for the development of a
stand-alone prototype. Object recognition and tracking algorithms were studied. In
this respect, already documented implementations were reviewed. The knowledge
acquired was later applied for the development of the prototype. After completion of
an executable prototype, various experiments to compare both systems were
conducted in a laboratory environment. Finally, the total costs of both systems were
assessed. For this purpose, active contracts of the supplier were studied and
analyzed.
This thesis enabled the client to gain initial experience in the field of artificial
intelligence. Algorithms for the recognition and tracking of objects, both proven and
new, form the central area of interest. Subsequently, the knowledge acquired was
applied for the development of an AI-based passenger counting system. Thereby it
became obvious that for the adaptation of such AI-based systems a reorganization
of the IT must first take place. New methodologies that allow fast and efficient
operation must be established and applied. Thus, potential challenges during
development and implementation can be efficiently handled. During a laboratory
experiment, the developed system was compared with the existing infrared
technology. It became apparent that AI-based passenger counting is feasible
although it does not yet pose a competition for the existing solution. Additionally, the
experiments revealed potential weaknesses of the current infrared system, which
were handed over to the customer for further investigation. Finally, the comparisons
in investment and maintenance costs highlighted the potential cost savings by
switching to AI-based passenger counting.
Studiengang: Business Information Technology (Bachelor)
Vertraulichkeit: öffentlich