I was invited to speak about the work I had done during my internship program at Daal navigator. I devised a "Traffic Engine" for building and maintaining a model of navigation graph traffic from a stream of user GPS data.
Interna is an event meant to convey internship experience from top students who have attended these programs to other interested students.
This presentation is my keynote for that speech.
Speaker keynote - Interna 2018 - Department of Computer Engineering at Sharif University of Technology
1.
2. Developing a
Traffic model for transport routes
based on user-base location data
By: Mohammadmahdi Faryabi
Details: Interna 2018
Thursday Aban 3rd 1397 (October 25th 2018)
Sharif University of technology
3. 13A GPS based Traffic Model
Interna 2018, Thursday Aban 3rd
Mohammadmahdi Faryabi
Presentation Scenario
•Map matching
•GPS model
•Deep learning model
3
•Introduction
•Problem definition
•Solution requirements
•Map modeling
•GPS data problems
4. 13A GPS based Traffic Model
Interna 2018, Thursday Aban 3rd
Mohammadmahdi Faryabi
Introduction
• Mohammadmahdi Faryabi
• Undergraduate student. CE @ SUT
• Traffic division R&D @ Daal co.
4
5. 13A GPS based Traffic Model
Interna 2018, Thursday Aban 3rd
Mohammadmahdi Faryabi
Problem definition
• Early days: using third party SDKs
• Google Maps API
• Need to have our own Traffic
information system.
• Change of business model in Google
Cloud platform
5
6. 13A GPS based Traffic Model
Interna 2018, Thursday Aban 3rd
Mohammadmahdi Faryabi
Solution requirements
• Positional data of our user base.
• Mathematical model for transportation routes.
• Relating positional data to model.
• A final model for traffic data of the map model.
6
7. 13A GPS based Traffic Model
Interna 2018, Thursday Aban 3rd
Mohammadmahdi Faryabi
Map modeling
• Too many alternatives:
• Hard field work to create the map.
• Use open source maps and
perform improvements.
• A business contract.
• Map needs maintenance:
• Automate via user feedback.
7
8. 13A GPS based Traffic Model
Interna 2018, Thursday Aban 3rd
Mohammadmahdi Faryabi
GPS data problems
• Many error factors:
• Decision on user activity type is not easy and is error
prone.
• GPS data accuracy is not always acceptable.
• GPS sample time is not reliable.
• Based on device time
• There are network latencies and disorders are
common.
• Hardware issues may cause bad GPS sample
behavior.
8
9. 13A GPS based Traffic Model
Interna 2018, Thursday Aban 3rd
Mohammadmahdi Faryabi
Map matching
• Assigning a position on the mathematical map
model to a GPS sample.
• A single point cannot be properly mapped.
• Matching is computationally expensive.
• Need to match many samples per second.
• Problem with routes with different altitudes.
• Problem with abnormal sample behavior.
9
10. 13A GPS based Traffic Model
Interna 2018, Thursday Aban 3rd
Mohammadmahdi Faryabi
GPS Model
• Using map matched GPS samples to maintain
and update a traffic model for map.
• Need to filter unwanted samples:
• Samples from users not in a vehicle.
• Samples from users with abnormal activity.
• Outlier samples.
• Need to differentiate between different users
as well.
10
11. 13A GPS based Traffic Model
Interna 2018, Thursday Aban 3rd
Mohammadmahdi Faryabi
GPS Model (cont’d)
• Need for a basis for our estimation of traffic.
• For every part of route in our model for different times
of the day, week and year.
• Need to update this basis estimation over time.
• Need to be able to swiftly diverge from this basis
temporarily in case of abnormal traffic events.
11
12. 13A GPS based Traffic Model
Interna 2018, Thursday Aban 3rd
Mohammadmahdi Faryabi
Deep learning model
• Problem of user base data being insufficient.
• Every one’s problem today.
• Need to measure traffic for all of the routes not only
for those which we have enough user data for.
• A deep neural network may be required.
12
13. 13A GPS based Traffic Model
Interna 2018, Thursday Aban 3rd
Mohammadmahdi Faryabi
Conclusion
• Map related APIs are among major requirements
for our startups and businesses.
• Alternative foreign solutions:
• Not tailored for our needs.
• Exceptionally expensive nowadays.
• There is still a whole lot more to be done.
13
14. 13A GPS based Traffic Model
Interna 2018, Thursday Aban 3rd
Mohammadmahdi Faryabi
Any Questions?
Thank You for your attention!
Editor's Notes
توسعهی سیستمی برای پیشبینی ترافیک معابر با استفاده از اطلاعات جغرافیایی کاربران
نعریف مساله و نیازمندیهای رسیدن به جواب و چالشهایی که در مسیر باهاش روبروه هستییم
هر کار متن باز احتمالا کیفیتش کمتر از ورژنهای مشابه تجاریش هست پس باید improvement انجام بشه
قرارداد تجاری و استفادهی نقشهی ایجاد شدهی دیگران
وارد کردن اطلاعات با اکتیویتی اشتباه پیامد بدی ممکنه داشته باشه مثلا روی بخشی از مدل که اطلاعات مکانی کمی داریم یک اطلاع اشتباه مثلا از یه کاربر که وایساده ممکنه باعث گزارش ترافیک بشه.
دقت دادهی gps ممکنه چند هزار متر در شعاع برسه. چنین دادهای عملا ارزش اطلاعاتی خاصی نداره برامون.
زمان ثبت دادهی gps اگه از ماهواره گرفته بشه ممکنه به خاطر تاخیرهایی که داریم خیلی بی دقت بشه اگر هم که از زمان دستگاه استفاده بشه ممکنه به خاطر اشتباه بودن زمان دستگاه دادهی دو روز پیش به عنوان دادهی امروز تلقی بشه!
تاخیرهای زمانی شبکه طبیعین و نمیشه کاریشون کرد.
در مورد سخت افزار خراب حتی زمان بندی خراب سخت افزار ممکنه باعث تولید سمپلهای gps ای بشه که خیلی عجیبن. م