As the host institute of “Sidewalk dataset construction project”, one of the “2019 AI learning dataset construction project” colluded by National Information Society Agency (NIA), Let me introduce the dataset construction examples.
<2019 “Sidewalk” Dataset Construction Project Host institute – Testworks>
Sidewalk Dataset Construction Project
Sidewalk dataset construction project is establishing dataset to detect various obstacles (such as cars, other walkers, streetlights and street trees) and damaged sidewalk surfaces for people with disabilities. This project is to create AI service development foundation for improving the mobility right of people with disabilities.
The sidewalk dataset created from this project reflects the domestic sidewalks different from the autonomous guided vehicle dataset. It has been committed to secure creativity and diversity as an open domestic dataset as there is no domestic and international parallel case in the subject and scale.
AI Ecosystem Formation & Social Value Creation
Testworks constructed Korea’s first sidewalk dataset suitable for the domestic environment and has laid the foundations to improve the movement right of people with disabilities. Testworks opened the dataset to the public for small and medium-sized businesses and startups with insufficient capital and resources to be able to cut down the cost in data collection and processing and contributed to related AI ecosystem formation.
Furthermore, during the dataset creation process, Testworks collaborated with KSCIA for the beneficiaries to be involved in the technical development steps. Disadvantaged groups in employment such as career-interrupted women, people with developmental disabilities and youth were hired and actively engaged. Testworks suggested a new employment example to the disadvantaged groups proactively corresponding with the ever-changing working environment in the 4th industrial revolution.
Sidewalk Dataset Construction Project = Short Term + Large Scale + High Quality
<Sidewalk Dataset Construction Project Performance and Achievement Index>
Testworks hired 56 new personnel during the six months of the sidewalk dataset construction project period. Testworks secured the necessary workforce in a short amount of time by cooperating with training centers for people with developmental disabilities, Dongbu Women Resources Development Center, Eunpyeong Women Resources Development Center, etc. Testworks ensured the quality of data by improving the professionality of new workers in data processing with systematical data processing and manager education.
Efficient Dataset Construction with “auto + manual” Package Processing & 3-step Review
With self-developed automation and developer-friendly web-based processing tools and 3-step review, Peer Review-Manager Review-Final Audit, in data processing procedure, Testworks received Pass in all 120 checklists of the final audit step and created the high-quality large scale dataset.
Testworks’ crowdsourcing platform aiworks and data processing management platform blackolive
Sidewalk Dataset Application Service
The dataset has been applied to various service creation such as road surface stability monitoring service and Korean food delivery robots for the delivery service as well as AI service development for public interest to enhance the mobility right of people with disabilities such as pavement danger warning system for people with visual impairment and safe route direction system for wheelchair users.
<Exemplary Services with Sidewalk AI Dataset Application>
Exemplary Services at Participating Institutes
KAIST RCV Lab developed an advanced object/distance recognition AI model for the pavement environment, Yungi Kim from SelectStar Inc., student, created a prototype of walkable district recognition models and DTWORESOURCE Corporation established an exemplary service for road surface safety monitoring system.
<Participating Institutes AI Model Exemplary Service>
Testworks Exemplary Service – Deep Learning-based Pedestrian Assisting System
Testworks developed an exemplary pedestrian assisting system smartphone application with sidewalk dataset and deep learning technology to detect objects in front, road surface, other surrounding obstacles and so on. This app is selected as an excellent application case of NIA AI learning data.
<Pedestrian System Structure>
<Source: Excellent Application Cases of NIA AI Learning Data>
The exemplary service above is selected as an excellent application case of NIA AI learning data.
Testworks aims at continuously creating social values with advanced technology and solving social problems.