Automated detection and counting of pedestrians on an urban roadside
Automated detection and counting of pedestrians on an urban roadside
Academic Level: Masters
Volume of 15 pages (4125 words)
This study of Automated Detection and Counting of Pedestrians and Bicyclists along an Urban
Roadway was undertaken as part of the Massachusetts Department of Transportation
(MassDOT) Research Program. This program is funded with Federal Highway Administration
(FHWA) Statewide Planning and Research (SPR) funds. Through this program, applied research
is conducted on topics of importance to the Commonwealth of Massachusetts transportation
As accommodations for pedestrians and bicyclists become an increasingly important part of
urban transportation planning, there is a growing need for accurate counts of these groups.
Pedestrian and bicyclist count data can be used for a variety of purposes; including intersection
planning, bicycle lane allocation, sidewalk design, and traffic light deployment, among others. It
has been shown that pedestrian and bicyclist injuries and deaths in urban areas can be
significantly reduced by the use of effective transportation infrastructure. Accurate pedestrian
and bicyclist counting methodologies serve as important resources for urban planners by
providing accurate data that can assist in transportation planning.
The counting of pedestrians and bicyclists has been an active research topic for over ten years.
To be useful, pedestrian and bicyclist counting systems must meet a series of criteria including
accuracy, ease of deployment and cost effectiveness. Current standards indicate a need for a
count accuracy of at least 85% over a period of hours in order for count data to be useful for
transportation planning. However, several limitations have made this level of accuracy difficult.
The use of human assistants to collect counts in real time is not only labor intensive, but can be
highly inaccurate. Additionally, early efforts to automate counting with low complexity
equipment, such as pressure sensors, were met with limited success due to inaccuracy. Another
issue has been one of privacy, as the most recent imaging-based counting systems require the
storage and post processing of image data to achieve accurate counts, which is a significant
individual privacy concern. Also, many current camera-based counting systems are quite
complex and require significant user expertise for proper operation. Luckily, recent
improvements in digital camera and imaging software technologies have made advances in
bicyclist and pedestrian counting more feasible.
In an effort to address the limitations of previous pedestrian and bicyclist counting systems, an
advanced camera-based system was developed for this project. The new system retasked an
existing traffic camera, the Autoscope Solo Terra, which is typically used to detect and count
motor vehicles at intersections and on highways, to focus on pedestrian and bicyclist counting.
For the project, a series of software enhancements were made to the equipment, in order to
optimize the detection of pedestrians and bicyclists both on a sidewalk, as well as in an adjacent
roadway bicycle lane. Using this retasked camera, all counts are determined in real time, using
software embedded within the camera and additional software algorithms implemented in an
attached personal computer. The key algorithm used to identify pedestrians considers the size of
an object located in an image zone in comparison to the known size of a pedestrian object.
Bicyclists are located by the presence of an object in a specific image zone for a period of time.
To increase accuracy, a second camera can be used to collect additional images, which may be
utilized to verify pedestrian counts from the first camera. The second camera uses a pedestrian
recognition technique based on the identification of a pedestrian’s head and shoulders. Multiple
identifications in an image frame indicate the presence of multiple pedestrians. The use of a
second camera is not needed for bicyclist detection since the single-camera approach is
The software interface for the developed system is easy to use. A transportation employee can
easily deploy the system by assembling it with simple tools and clicking on several icons on a
personal computer desktop. For advanced use, software can be adjusted to become more or less
sensitive to individual pedestrians and bicyclists. Our testing has indicated that high count
accuracy can be achieved for a range of settings. Two deployment platforms have been
constructed and tested for the camera-based system and an initial platform based on a stepladder
was used to generate the results documented in this report. In the latter stages of the project, the
stepladder was replaced by a trailer which can easily be attached to a transportation department
motor vehicle. This ruggedized system is available for immediate deployment by transportation
department employees in areas where pedestrian and bicyclist counts are needed. In both cases,
the retasked traffic camera is mounted on an extendable pole and pointed perpendicularly to the
flow of pedestrian and bicyclist traffic.
ES.3 Results and Conclusions
To verify the results of the system, a series of experiments were performed using the integrated
experimental setup at the University of Massachusetts, Amherst, and in downtown Boston.
Pedestrian traffic on an enclosed footbridge, an open pedestrian path and an urban sidewalk were
evaluated over a span of more than ten minutes per experiment. Real time pedestrian counts
taken using the zone-based method alone exhibited better than 85% accuracy while the combined
detection zone/histogram of oriented gradients (HoG) approach consistently approached 90%
accuracy. Tests were performed for a variety of pedestrian foot-traffic densities. Similar accuracy
was determined for bicyclist counting.
The approach also successfully counts pedestrians moving in opposite directions on the same
sidewalk at the same rate. Bicyclist counting with accuracy similar to pedestrian counting is
limited to unidirectional flow for bicyclists operating in a lane adjacent to an urban sidewalk.
The accuracy of our approach is limited by sensitivity to shadows and strong bursts of sunshine,
but future enhancements in imaging software may help to address these issues.
In conclusion, this effort to build a practical, easy-to-use, and precise pedestrian and bicyclist
counting system has resulted in a deployable system which is highly accurate. The system has
been tested on an urban roadside for an extended period of time to determine its long-term
effectiveness. Future work will involve making the system more robust to solar glare, shadows
and darkness. Further testing in a variety of weather conditions is also desirable. As a result of
this work, it is recommended that a pilot project be established that all allows for the extensive
collection of bicyclist and pedestrian data in a variety of real world urban roadway environments.
Given the immediate need for this data and the availability of the functional prototype, the
collection of this data can have significant short and long term benefits forurban transportation planning.