Camera Analytics Part 2: A(Eye)

ADOT Systems Technology

Kendra Johnson & Steven Cheshko

Agenda

• Recap of Part 1
• Bosch AI
• TrafficVision
• Comparison

​

​

​

​

2

Source: Pixabay

Recap of Part 1

Analytics to help Traffic Operations Center (TOC)

​

Can be done through system (centralized) or device (edge)

​

Existing Bosch cameras

​

​

​

​

​

3

Image from

Configuration Manager

Recap of Part 1

Strengths

• Picking up vehicles
• General trajectories

​

Challenges

• Small objects
• Far objects
• Stopped objects
• Clustered objects

​

​

​

​

​

Opportunities

• Drops in speeds
• “Loitering” objects

​

​

​

​

​

​

4

Bosch AI

More processing power

​

Pre-trained to focus more on transportation items

​

Edge device

​

Single unit for testing

​

​

​

5

Bosch AI

Improvements

• Object recognition
• Counts/Classification
• Congestion
• Vehicle movement
• Stopped vehicles
• Wrong-way false calls

​

​

​

​

​

​

​

​

6

7

Bosch AI

Struggles

• Speeds
• Average was lower
• Individual decreased with distance
• Debris
• Not a “transportation” class

​

​

​

​

​

​

​

​

8

Introduction to TrafficVision

• Real-time Anomaly and Incident Detection Data
• Collects Data in Non-Intrusive Way
• No New Infrastructure
• Use Existing Camera Feeds
• Streams at 240 PPI
• Low Resolution
• Collects Traffic Data
• Volume, Average Speed, and Classification Data
• Lane-by-Lane Data

9

Traffic Vision Pilot

• Central System - Hardware Agnostic
• Dates: 10/8/2024 through 3/31/2025
• Live Stream 20 Cameras
• 18 CCTV
• 2 WWD Thermal Cameras
• 1,824 events
• 263 with Wrong Way Thermal Cameras
• 1,561 with CCTV Camera

​

​

​

​

10

Locations

11

• SR 202
• 0012W SR 202 E of 24th St
• 0496E SR 202 Dobson Rd
• 0455E SR 202 Cooper Rd (10/8/24 - 2/6/25)
• 0085W SR 202 McClintock Dr (10/8/24 - 1/31/25
• 0796S SR 202 Van Buren St (10/8/24 - 11/30/24)
• Others
• 1101S SR 303 Northern Ave
• 1789E US 60 Alma School Rd
• 1729E US 60 W of Mill Ave (2/10/25 - 3/31/25)
• 2013S I-17 S of Thomas Rd

​

• I-10
• 1546W I-10 E of Broadway Rd
• 1456M I-10 7th St
• 1332W I-10 99th Ave
• 1437M I-10 19th Ave
• 1453E I-10 Tunnel E/B (12/4/24 - 3/31/25)
• 1564W I-10 E of Baseline Rd
• 2562E I-10 Grant St (Tucson)
• SR 101
• 0029N SR 101 Thomas Rd
• 0250E SR 101 W of 7th Ave
• 0339E SR 101 Scottsdale Rd (2/11/25 - 3/31/25)
• 0542N SR 101 N of Southern Ave
• 0517N SR 101 N SR 202 Red Mtn. N.W. Ramp

​

Daily Count of Alerts

12

Average is 10.5 Alerts per Day for 20 Cameras

Category of Alerts

• There are 7 Types:
• Stopped Vehicle/Debris
• Congestion
• Low Visibility
• Pedestrian
• Slow Speeds
• Fast Speeds
• Wrong Way Driver
• Alerts are Customizable

13

Correct vs Incorrect Alerts

• Focused on the 1,561 CCTV Alerts
• Correct Alerts include:
• Car Stopped at Intersection
• Car Stopped at Ramp Meter
• Stopped Construction Vehicles
• Traffic
• Car in Shoulder
• Incorrect Alerts include:
• Camera Pointed at Sky (Low Visibility Alert)
• Car Identified as Pedestrian (Pedestrian Alert)
• Camera Zoomed In (Slow Speeds Alerts)

14

Correct But Unnecessary

15

Median Cameras - 19th Ave

16

False Alerts

17

Obvious Misses

18

Field Testing

19

AI Verification Analysis

20

Alerts Not AI Verified: 938

Incorrect: 323

Correct: 615

AI Verified Alerts: 623

Incorrect: 67

Correct: 556

21

Categorization of Alerts for Analysis

22

Are the Alerts Helpful?

“Incidents” and “Car in Shoulder” alerts were classified as “Helpful” while the rest of alerts were classified as “Not Helpful”

23

24

25

26

27

Incident Timestamps

• Timestamps of:
• TrafficVision Alerts
• TOC Spillman CAD Entry
• DPS Calls
• Crash Record
• 18 Incidents
• Caveats:
• Takes a few minutes for TOC to verify call and input incident into CAD delaying the recorded timestamp

28

29

Incident Timestamps Statistics

• 18 Total Incidents
• 17 Incidents TOC Recorded
• All After TrafficVision Alert
• 13 Incidents DPS Recorded
• 6 Before TrafficVision Alert
• 6 After TrafficVision Alert
• 1 Same Time as TrafficVision Alert

​

30

Wrong Way Driver Analysis - 64th St

31

Wrong Way Driver Analysis - Maryland

32

Wrong Way Driver Analysis - Maryland

33

Events Only TV Got

Events Only Flux/Cam Got

99 Total Events

Wrong Way Driver Analysis - Van Buren

34

Wrong Way Driver Analysis - Van Buren

35

Events Only TV Got

Events Only Flux/Cam Got

128 Total Events

36

Orange - RADS

Blue - TrafficVision

37

Orange - RADS

Blue - TrafficVision

Red - INRIX

Recommendations for TrafficVision

• Avoid Median Cameras
• Avoid Rush Hour Alerts
• Turn Off Algorithm During Rush Hour
• Avoid Cameras in High Traffic Areas
• Avoid Construction Areas
• Avoid Cameras that Overlook/Focuses Intersections
• Outskirt, Low Traffic Cameras
• Use During Off-Peak Times
• TrafficVision’s Traffic Data is accurate but only in the Right Circumstances

38

TrafficVision Vs Bosch

• Bosch
• “Free” - included with cost of the camera
• Dependent on the camera
• No existing centralized dashboard
• Alerts only by text or email; *claim possible Cameleon integration
• Traffic Vision
• Subscription fee
• Does not require specific or advanced cameras
• Has existing centralized dashboard
• Text or email alerts available; *claim possible Cameleon integration

*Integration work was outside the scope of testing

​

39

TrafficVision Vs Bosch

• Bosch analytics set up on 3 cameras of TrafficVision pilot
• SR 101 Thomas Rd (AI)
• 10/20/2024 - 4/15/2025
• SR 101 Southern Ave and I-10 99th Ave (Standard)
• 3/5/2025 - 4/15/2025

40

TrafficVision Vs Bosch

41

Events Only TV Got

Events Only Bosch Got

118 Total Events

Of the 21, TV alerted to 12 of them before Bosch

Conclusions

Neither system caught everything, but they move the needle

Focus on the right circumstances

• Pros and cons of each approach
• Try to reduce amount of false positives
• Dependent on presets, especially for Bosch
• May be more useful in lower traffic areas
• Adjust for rush hour
• Neither really worked well for debris
• Could supplement count stations
• Useful for stopped vehicles

​

​

​

​

​

42