Hosted by: The Data Science for Social Good (DSSG) program at the University of British Columbia
Sponsor: Cedar Academy Society - VanCom Project
Time: May 2021 - August 2021
Description: Public sector and academic communities have been using mobility and traffic data as a proxy measurement for a variety of social topics, from GDP prediction and economic development to greenhouse gas emissions and environmental impact. One method to measure mobility and acquire traffic data is through the analysis of pictures and footage from traffic cameras installed at fixed locations (in urban and rural areas). More often than not, the cameras are installed near locations with heavy traffic, and this introduces sampling bias in the observed data. This leads to a biased dataset and overexaggerates nearby mobility levels due to “preferential sampling.” This project seeks to correct this preferential sampling and develop an algorithm to better model mobility levels while accounting for the bias in the data set.
Request for download link at data@pwfh.org, for data items with this
Click to download a high-level VanCom Mobility Data Users' Guide.
Data from Cedar Academy Society:
-
Time: December 1-31, 2020
Assets: 364 location-based Assets in the City of Surrey, British Columbia, Canada
Description: The file contains full-YOLOv3-extracted information from raw static image files. For that reason, please ignore the "MOV index" part (this is for clients who need aggregated and scaled data) in the Users' Guide.
-
JPEG files from two camera stations have been pulled and stored below for your reference
Time: December 1-31, 2020
Assets: 2 location-based Assets in the City of Surrey
- 104 Ave And 140 St
- 104 Ave And City Hall Driveway
104 Ave And 140 St(station name: enc_104_140_cam1) | 104 Ave And City Hall Driveway(station name: enc_104egress_cityhall_cam1) |
---|---|
-
Time: October 10-16, 2020
Assets: 364 location-based Assets in the City of Surrey, British Columbia, Canada
-
Time: May 2-3, 2020
Asset: 1 location-based Asset: 104 Ave And 140 St
Data and tools from other sources:
-
Interesting datasets on the City of Surrey here
and Surrey 2015 Traffic Counts dataset
Not all traffic signal location will have cameras
-
StatCan boundary data (.shp file):
Road network in Surrey and from DataBC
-
Geocoding tools:
The following links that can give you inspiration of how to get information about the "features" of an intersection
-
Mobility data in Real Estate context and Canadian Economics Society Annual Meeting 2021 presentation
-
GHG emission indexingThe idea is to project vehicle type/make recognition to Green House Gas emission. A starting point can be car recognition mechanism from image/video files, e.g., here on Github
and a Standford car dataset here.
The challenge in this case is the low resolution of raw image/video files that makes ID of vehicles logo and shape of headlights impossible.
-
Crime Prevention [7]
The idea is a Minority Report kind of mechanism
-
Economic Recovery post Pandemic [4]
Mobility and Engagement Index by the Federal Reserve Bank of Dallas
The Impact of COVID-19 on Small Business Dynamics and Employment
Bloomberg article: High-Frequency Data Prove Their Staying Power With Fed’s Buy-In
-
Weather influence on Mobility
-
Bike routes monitoring
Cam and Shared Traffic | Cam and Bike Lanes |
---|---|
-
Bias correction
-
Metadata
-
Better Object Detection
-
Crowd Counting / Crowd Classifier [1]
-
Rolling window
-
Background Subtraction with OpenCV and a comprehensive intro here
-
YOLOv3 --> YOLOv4 --> YOLOv5, PAFNet, Rotate Anchor, PaddleDetection
-
Image pre-processing
Gamma correction or histogram transformations. Some examples are here:
-
CV model trained on traffic camera datasets [6]
-
Try Yolov5
Linux, inference with detect.py
# clone github repo git clone https://github.com/ultralytics/yolov5.git # clear requirements and download pretrained model weights cd yolov5 pip install -r requirements.txt python detect.py --weights yolov5s.pt python detect.py --weights yolov5m.pt python detect.py --weights yolov5l.pt python detect.py --weights yolov5x.pt # Try Yolov5 on static image files python detect.py --source 2020-12-06-23-22-42-enc_104_140_cam1.jpg --weights yolov5s.pt --project infer_yolov5s_2 python detect.py --source 2020-12-06-23-22-42-enc_104_140_cam1.jpg --weights yolov5x.pt --project infer_yolov5s_3
- Try PaddleSeg for Semantic Segmentation
AutoNue Challenge 1st Prize Winner at CVPR 2021
-
-
Edge Computing
-
Using edge device to extract traffic insights [3]
for example, Papers with Code - Mobile Video Object Detection with Temporally-Aware Feature Maps
-
See Microsoft Project Rocket for Vision Zero -related projects insight
-
-
Hardware
- ESP32 and ESP32-CAM
-
Scalable Object Detection Pipeline
- NVIDIA, Deploying a Scalable Object Detection Inference Pipeline Blog: Intro
- NVIDIA, Deploying a Scalable Object Detection Inference Pipeline Blog: The Inferencing Process
- VanCom partnership with industry leader SafeGraph: Placekey + VanCom: Combining Anonymized Traffic Activity Datasets and POI Data for Advanced Analytics
- NeurIPS 2021’s competition Traffic4cast
- Kaggle's Android smartphones high accuracy GNSS datasets
- Standford car (type/make) dataset here
- UA-DETRAC dataset (detection/tracking), with research paper here [8]
- CVOnline - Compendium of Computer Vision
- NVidia AI CITY CHALLENGE
[1] Understanding Traffic Density from Large-Scale Web Camera Data, Shanghang Zhang, Guanhang Wu, João P. Costeira, José M. F. Moura, arXiv:1703.05868 [cs.CV]
[2] A general theory for preferential sampling in environmental networks, Watson, J, V. Zidek, J, Shaddick, G, Annals of Applied Statistics, 2019, 2662-2700
[3] Object Counting on Low Quality Images: A Case Study of Near Real-Time Traffic Monitoring, Jean-Francois Rajotte, Martin Sotir, Cedric Noiseux, Louis-Philippe Noel, Thomas Bertiere, IEEE Xplore: 17, January 2019
[4] Mobility and Engagement Following the SARS-Cov-2 Outbreak, the Federal Reserve Bank of Dallas
[5] Leveraging Administrative Data for Bias Audits: Assessing Disparate Coverage with Mobility Data for COVID-19 Policy, Amanda Coston, Neel Guha, Derek Ouyang, Lisa Lu, Alexandra Chouldechova, Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency. pp. 173-184, arXiv:2011.07194 [stat.AP]
[6] STREETS: A Novel Camera Network Dataset for Traffic Flow, Advances in Neural Information Processing Systems 32 (NeurIPS 2019)
[7] Leveraging Mobility Flows from Location Technology Platforms to Test Crime Pattern Theory in Large Cities, Cristina Kadar, Stefan Feuerriegel, Anastasios Noulas, Cecilia Mascolo, arXiv:2004.08263 [cs.CY]
[8] UA-DETRAC: A New Benchmark and Protocol for Multi-Object Detection and Tracking, Longyin Wen, Dawei Du, Zhaowei Cai, Zhen Lei, Ming-Ching Chang, Honggang Qi, Jongwoo Lim, Ming-Hsuan Yang, Siwei Lyu, 2020, arXiv:1511.04136 [cs.CV]
[9] Urban flows prediction from spatial-temporal data using machine learning: A survey, arXiv:1908.10218 [cs.LG]