Computer vision for pedestrian accessibility from street-level imagery - BC-1027

Genre de projet: Recherche
Discipline(s) souhaitée(s): Génie - informatique / électrique, Génie, Informatique, Sciences mathématiques, Ressources et gestion environnementales, Sciences naturelles
Entreprise: Ponder Technologies Inc.
Durée du projet: Flexible
Date souhaitée de début: Dès que possible
Langue exigée: Anglais
Emplacement(s): Vancouver, BC, Canada; Burnaby, BC, Canada; Greater Vancouver Regional District, BC, Canada; Canada
Nombre de postes: 1
Niveau de scolarité désiré: MaîtriseDoctorat
Ouvert aux candidatures de personnes inscrites à un établissement à l’extérieur du Canada: Yes

Au sujet de l’entreprise: 

Ponder Technologies Inc. is an early-stage Vancouver based company building CAW, a civic field-capture platform. Compensated “spotters” walk municipally-ranked routes and capture continuous, current, street-level imagery of failing public infrastructure: cracked sidewalks, inaccessible curb ramps, damaged transit stops, overgrown trees, and other public infrastructure that may be hazardous or cause accessibility and liability concerns. This data feeds a condition-analysis engine and a living map for different stakeholders: municipalities for infrastructure upkeep, sanitation, and barrier-removal obligations under the Accessible British Columbia Act; utility providers for line-clearance risk; and transit operators for the condition of stops and pedestrian access.
The platform grew out of a property/home maintenance product which pivoted towards public-good civic infrastructure and accessibility data platform after market validation feedback received from a business accelerator program. The founder is a solo technical founder with an Avionics background currently working in the aviation industry. Ponder is now seeking to assemble a research team to build the machine-learning core of the platform properly.

Veuillez décrire le projet.: 

Goal: Build the machine-learning core of CAW: a computer-vision system that automatically detects and classifies accessibility features and defects in the pedestrian environment: curb ramps (and their compliance), sidewalk cracking/heaving and surface problems, obstructions, missing infrastructure, and related barriers directly from the continuous, on-foot, GPS-tagged street-level imagery CAW captures, and localizes each finding on a map with a severity and evidence class.
Why it matters: Existing academic and municipal accessibility mapping is constrained on one side by slow, small-sample manual ground-truthing, and on the other by a reliance on existing online map imagery that is captured from vehicles and refreshed on the imagery provider's schedule, so much of the pedestrian realm and its current condition goes uncaptured. CAW supplies fresh, owned, continuously-updated imagery collected on foot; this project builds the ML layer that turns that imagery into structured, validated accessibility data at scale.
Intern tasks: Assemble and annotate a labelled dataset from CAW imagery; develop and evaluate object-detection / semantic-segmentation models for sidewalk and curb-ramp condition; geo-reference detections and quantify accuracy against ground-truth audits; document a reproducible pipeline. This directly advances UN SDG 11 (inclusive, accessible public space) and supports municipal obligations under the Accessible British Columbia Act.

Expertise ou compétences exigées: 

Computer vision and deep learning (object detection, semantic segmentation); Python and a modern ML stack (PyTorch or TensorFlow). Strong data-handling: dataset assembly, annotation workflows, and model evaluation. Working knowledge of geospatial / GIS concepts (coordinate systems, geo-referencing imagery, mapping detections) is highly desirable.
Desirable: experience with street-level or aerial imagery, edge/mobile inference, or human-centred / community-based research. An interest in urban accessibility, gerontology, or the built environment is a strong asset given the applied, public-good nature of the work.
Assets available to the intern: an early stage live capture platform generating proprietary imagery and telemetry; an existing basic production data pipeline.