Now You See Me, Now You Don’t: Automatic Pedestrian Detection at Signalized Intersections and Midblock Crossings

In 2018, Florida was ranked as having the third-highest pedestrian fatality rate in the United States, according to the National Highway Traffic Safety Administration. Pedestrian safety is a nationwide issue, and it is a top priority in Florida being tackled by adding more safety features and improving facilities. Unfortunately, finding solutions is not as simple as safety engineers would like.

Take, for example, the pedestrian push button installed at many signalized intersections:  it sends a call to the traffic signal’s controller and prompts a “walk” signal to cross the road to create a safe crossing.  Yet, in 2013, according to the Pedestrian Safety Guide and Countermeasure Selection System, 40 to 50 percent of pedestrians do not use or have difficulty using the pedestrian push button. There are a number of reasons why the buttons do not get pushed. Some pedestrians mistakenly assume the button has been pressed already by another pedestrian. Others assume the walk signal is automatically given for every signalized intersection. One increasingly relevant reason is sanitation; given COVID-19 and the risk of infection, fewer people are willing to touch public surfaces.  This last reality begs the question: why does the button need to be pushed at all?

If cars can be automatically detected at signalized intersections, why not pedestrians? From safety concerns like pedestrians crossing without being prompted by the pedestrian walk signal to more practical maintenance issues like stuck buttons causing unnecessary vehicle delay, the existing pedestrian push-button technology has its issues. To that end, FDOT sought a passive, or automatic, pedestrian detection technology for signalized intersections and mid-block crosswalks that could improve pedestrian safety.

A Long Time Coming

Automated pedestrian detection identifies humans and prompts a walk signal without the pedestrians needing to press a button. It has the potential to be used not only at signalized intersections, but also where other pedestrian safety features are implemented – like rectangular rapid flashing beacons (RRFB) and high-intensity activated crosswalk (HAWK) signals.

FDOT Project Manager Alan El-Urfali says the discussion about passive detection has been a long time coming. “The earliest occurrence of this technology I know of is from a 1997 article for passive detection, which was related to the development of the HAWK, or pedestrian-hybrid beacon. This was before the HAWK signal was approved in the MUTCD and came out in 2009.” At that time, however, the technology was limited and its effectiveness poor: “There were false calls everywhere.”

Now several detection systems are available, with varying levels of accuracy, reliability, and cost. They include infrared, microwave, and thermal sensors; pressure mats; and computer-assisted video. Determining which system or systems are best suited to the needs of Florida became the challenge. FDOT engaged a research team including Principal Investigator (PI) Dr. Pei-Sung Lin, Co-PIs Dr. Achilleas Kourtellis and Dr. Zhenyu Wang, and research faculty Dr. Cong Chen with the Center for Urban Transportation Research Center (CUTR) at the University of South Florida (USF). They first developed a list of criteria when reviewing existing detection systems. Dr. Lin explains, “We looked at different vendors, some from different fields of work but with the potential to be applied to our research, and some that were already available and on the market. We reviewed 12 technologies, then eventually narrowed down to three vendors to fully test.”

Testing Complexities

The three selected vendors were first tested in a controlled lab setting to obtain data and learn more about how the systems function. Then, the systems were installed and tested on Tampa’s University of South Florida campus at a signalized intersection and two midblock crosswalks.

Dr. Lin explains there are a lot of complexities to consider when testing detection in the field. “Consider the direction in which you are traveling. Are you exiting the crosswalk instead of entering? You don’t need to be detected when you are finished crossing the road. The detection system should be able to detect and recognize which direction you are going. Do you want to trigger the walk signal or not?”

Another complexity to test was the ability to deactivate, or cancel, a call that was activated using the pedestrian button. This is necessary in situations where a pedestrian may wait to cross one leg of the street but then decides to first cross the adjacent leg, or where a pedestrian sees a gap in traffic and crosses without the walk indication. Dr. Lin describes, “Once you push a ped button, a call has been made to the traffic signal controller. But if we can see a person press the button, then leave, would the system be able to cancel that call? The walking indication is not needed anymore, yet the major street has to wait if the pedestrian call from the side street is not cancelled. It causes unnecessary delays for vehicles.”

Different clothing, movements, and objects under various weather conditions and times of day were used to assess the systems in the field. Pedestrians wearing raincoats, holding umbrellas, and sitting in wheelchairs tested the systems. Dr. Lin asks, “When one stands still, how do you know it’s not a fixed object? How can you recognize a pedestrian using an umbrella? How do you know it’s a human being?”

The research team found that the thermal machine vision system worked best at all testing locations, having the highest detection accuracy when desired and lowest false detection rate when not needed. When tested at the midblock crossings, the system accurately detected desired movements 92% of the time with a 2% false positive rate. At the signalized intersection, the system accurately detected pedestrians 94% of the time and activated calls 90% of the time.

Additionally, the system accurately detected disappearing pedestrians (that is, pedestrians who arrive at the crossing but depart before receiving the walk indication) 98% of the time and removed unneeded pedestrian calls 97% of the time. Dr. Lin explains that the detection system would wait to place a formal call to the controller until the moment the traffic signal was prepared to act on the information. “Here’s the hint. When a pedestrian is detected, a pedestrian call is held, but does not actually go through yet. Then, if the pedestrian is still in the waiting area by the end of previous signal phases for vehicles, the pedestrian call will go through and the walk signal will be displayed.” Dr. Lin further observes, “If a pedestrian disappears before the time for actually placing the pedestrian call in a traffic controller, the call on hold will be automatically released, so the pedestrian call will be not placed.”  This capability of only placing a call when it is needed ensures the intersection operates efficiently for both pedestrians and vehicles. 

The Future is Smart

The thermal machine vision system is promising for enhancing pedestrian safety and reducing vehicular delay at signalized intersections and midblock crossings. “It has not been used widely yet, and is expensive” Alan El-Urfali observes, but crashes are expensive too, and sometimes result in lost lives, so risk-based analysis is being used to determine where this technology should be implemented. “Ultimately, we want both passive and active ways to measure pedestrian detection. We don’t want to provide a false sense of security with only automated detection.”

Dr. Lin reflects on his research, saying, “People recognize the need for this technology. They know it’s achievable. This study demonstrates to DOTs and local agencies that better detection performance can be achieved.” This innovative technology is directly applicable not only to Florida but other states. Automated detection systems may be exactly what the nation needs to prioritize pedestrian safety and at the same time to reduce unnecessary vehicle delay.

Further Reading

BDV25-977-44 Integration of a Robust Automated Pedestrian Detection System for Signalized Intersections
Final Report | Summary