Traffic Solutions How to Test FCD Monitoring Solutions: Performance of Cellular-Based Vs. GPS-based systems
About Cellint Israel Based, office in the US Main products NetEyes for quality of RF networks TrafficSense for traffic monitoring (cellular-based innovative solution) State wide and metro wide deployments in the Israel, US and Europe (thousands of miles deployed) Granted patents in the US and Europe, pending patents in the rest of the world Company owners include Dr. Andrew Viterbi, Cofounder of Qualcomm and the inventor of the Viterbi code (i.e. CDMA/UMTS)
Customers/Partners info 24
Best Traffic Monitoring System Vs. the Worst Traffic Monitoring System Typical chart of speed over road segment with rush hour slowdown, by a good detection system will look as follows: Data from a broken traffic monitoring system, which only reports free flow, will look as follows: 4
Conclusion: Bad systems Can Hide Behind Statistics Observations: 1. The average daily error rate can be less than 5% for: A system that doesn t detect slowdowns at all A system detecting slowdowns with 50 minutes delay! 2. Sporadic FCD test drives: Will generate very little statistics for congested times Can t tell you how fast your system detected the slowdown Conclusions: 1. Average daily speed error should not be considered a parameter for evaluating traffic monitoring systems 2. FCD test drives is not a sufficient way to evaluate system 5performance
Same Data: Different Perception of Good and Bad Observations 1: Data Ok on average, but not always reliable Observations 2: Data VERY reliable, excellent slowdowns detection, no false alarms! Conclusion: wrong statistics can kill good systems 6
Testing Performance of FCD systems 7 What do we test? We don t need a system to detect free flow, so focus should be on slowdown detection Slowdown Detection Latency: Can only be tested by road sensors, since floating drive test can only randomly detect a slowdown when it starts Less than 8 minutes average latency can be provided on highways and major arterials Travel Time Measurements During Speed Fluctuations Must be conducted over short segments, otherwise traffic conditions can change significantly during a single measurement Must have statistical significance, since travel times may vary between two cars by 300% for the same traffic conditions (same road segment/ same time)
History Non intrusive FCD solutions seem to be the ultimate solution for future traffic data collection Until 2 years ago, all independent evaluations demonstrated relatively poor results: US Netherlands Italy Spain 8
9 The reasons for problematic performance GPS Fleet Tracking Only few samples per day per road section, even for most congested urban highways Long time between location samples Fleets tend to avoid congestions, if possible, so detection doesn t provide proper sampling As a results GPS aggregators provide predictive data based on historical patterns, not real time traffic information Cellular-based solutions If location is based on theoretical triangulation it s not accurate due to multi-path (blocking/reflecting elements)
The Change During the last 2 years the breakthrough, we were all waiting for, occurred: 5 independent evaluations, by different DOTs, show success of Cellint s TrafficSense system, which provides: Similar performance to road sensors in detecting slowdowns Best possible travel time measurement Similar performance to road sensors in measuring local speed In the following slides, we will review some of these test methods and the performance that can be generated from c cellular based systems 10
Atlanta Project Independent Evaluation TrafficSense provides area-wide coverage for construction zone management: 11 Most sensors over GA400, one of the busiest corridors in Atlanta, were disconnected due to construction TrafficSense provides data, since last year, over the highway and adjustment arterials Performance evaluation conducted by Georgia Department of Transportation Major findings: TrafficSense speed over short segments (250 meters) matched the sensors speed very well in all speed ranges TrafficSense travel time was highly accurate, even during the most congested times
12 Atlanta: TrafficSense Data is Incorporated for GA400 on Georgia DOT Official Website
Atlanta Project Independent Evaluation Speed Range (mph) Mean Difference % 20-30 3.82% 30-40 7.13% 40-50 6.65% 50-60 2.63% 60-70 -8.88% 13
KC Project Independent Evaluation System was deployed over entire corridor in less than 2 months TrafficSense data was compared to inductive loops data of SCOUT (KC traffic management center) Independent evaluation by Kansas Department of Transportation shows: 14 5.9 minutes average delay in detecting slowdowns Less than 5 mph speed difference in all ranges Quoting the pilot report: TrafficSense data clearly reflects traffic conditions very well The Cellint system successfully proved the viability of the technology for traffic applications
KC Project Evaluation Method [1] TrafficSense speed comparison with road sensors Magenta: Sensor 69 I-435 WB at US-69 Blue: TrafficSense 15 Speed changes are detected immediately (similar to sensors)
KC Project Evaluation Method [2] TrafficSense speed comparisons with road sensors Magenta: Sensor 69 I-435 WB at US-69 Blue: TrafficSense 16 Average speed difference over 5 days - 2.76 mph If no slowdowns were detected it would be 4.5 mph
KC Project Evaluation Method [3] TrafficSense speed comparisons with road sensors Magenta: Sensor 136 I-435 EB at Wornall Road Blue: TrafficSense 17 TrafficSense had significantly fewer false slowdown detections than the Kansas City sensors
Kansas City: Deployment Following a Successful Pilot 18 0-30 MPH: RED 30-45MPH: YELLOW 45+ MPH: GREEN
Tel-Aviv: Independent Evaluation of an Israeli DOT Agency 19 Two competing cellular-based solutions failed this test
9:36 8:24 Tel-Aviv Independent Evaluation [2] TrafficSense Travel-time Comparisons With Road Sensors and Test Drives (Drives conducted by the Israeli DOT) Minutes 7:12 6:00 4:48 3:36 2:24 1:12 Test Drive Road Sensors TrafficSense 0:00 14:40 pm 15:08 pm 15:54 pm Conclusions: Sensors are less accurate during speed fluctuations 20 TrafficSense travel-time average difference: 9.3%
Project in Sweden Deployment over Southern of Sweden for the Swedish Road Administration Cellint s TrafficSense system was selected following a competitive tender for Mobile Phone Traffic Monitoring system Metro wide deployment in the Skåne Region covers freeways and trunk road, including city streets in Malmö and Lund Successful evaluation of TrafficSense s traffic data has lad to project expansion last month
Project in Sweden
Israeli National Road Company: Project Evaluation Tested 3 cellular-based systems over Highway 1 and Arterial 44 Validated data from the independent evaluation demonstrated 1 minute latency for TrafficSense in detecting slowdowns as compared to road sensors, at the location of the sensors TrafficSense proved to be better than other solutions 23
24 Israeli National Road Company: Project Evaluation Interface
TrafficSense Slowdown Detection: A Delay of Only a Few Minutes 25 20 15 Operator's Subscriber Penetration (%) 10 5 0 Tel-Aviv Atlanta Kansas City Springfield MO Detection Latency Over Major HWs (minutes) 25
TrafficSense Unique Advantage All cellular-based technologies except TrafficSense use cell sites location to calculate vehicles location This theoretical calculations (either based on some type of triangulation or on cell sector statistics) has a lot of inherent in-accuracies TrafficSense doesn t require the location of the cell sites at all, but rather uses ground-truth reference data as a location reference. 26 This enables TrafficSense to achieve orders of magnitude better location accuracy and slowdown detection latency also in highly dense urban areas
Limitations of All FCD Monitoring Systems Not enough data for accurate detection during late night hours 27 But less than 1% of congestion events occur at night No accurate volume counting Accurate volumes are not critical for traffic control if you have local speeds, slowdown detection and travel times One can get good statistical estimates over time for road planning from the cellular system Can t differentiate between lanes But once a lane is separated by terrain it can be differentiated due to a different cellular signature (only TrafficSense) HOV lanes can be accommodated by using supporting systems
Limitations of GPS Based Solutions Only few samples per day per road section, even for most congested urban highways: Due to communication cost (sending data to central location) and privacy issues Long time between location samples Since by definition fleet tracking doesn t require location data every few minutes Fleets tend to avoid congestions, if possible, so they don t provide proper sampling when highly needed And taxi drivers tend to drive differently than normal traffic (usually more aggressive..) As a results GPS aggregators provide predictive data based on historical patterns, not real time traffic information Providing estimated information only for recurring rush hour slowdowns Even if non-recurring slowdown is sampled, it might be 28detected hours after it started
www.cellint.com TrafficSense Most Cost Effective Solution for Road Management The Only Validated FCD Solution for Accurate, Real Time Traffic Information 29 Cellint Traffic Solutions www.cellint.com info@cellint.com US 973-867-5595 Israel +972-524-77-33-77
Q&A: Smart Use of Cellular-Based Monitoring Systems 30 Would you pay orders of magnitude more on traditional systems for a metro-wide deployment just due to the night time detection limitation of cellular-based systems? Probably not Do you really need volume counting every half mile for real time road management purposes? Probably not. However a measurement at least in one point between each two junctions is useful for road planning How can I ensure proper performance and not get 50-minutes detection delays for slowdowns on major highways? Talk to colleagues from other States DOTs who conducted independent objective studies and who checked slowdown detection latency, not only average daily speeds
TrafficSense System Highlights Significantly Lower Cost Than Traditional Systems, Providing: Fast slowdown detection like road sensors Most accurate travel time Plug and Play System Modular system, super short deployment time Flexible to road changes and constructions Virtual Sensors in Small Intervals (every 250 meters in urban area) provide Both : Travel time updated to the minute Very accurate and immediate incident alerts Local traffic speed measurement similar to road traffic sensors 31