Pedestrian traffic monitoring is in its infancy, and the volatility of pedestrian traffic creates a need for guidance on site selection in traffic monitoring programs. A robust knowledge base surrounding pedestrian traffic patterns and the degree to which a single counting station is representative of a larger area are essential in developing an accurate program for estimating pedestrian traffic volumes. This research analysed long term hourly data from automated pedestrian counting devices on four consecutive blocks along an entertainment area corridor to determine the shifts in temporal pedestrian traffic characteristics and volumes along a corridor. Features of the built environment were identified that can aid in estimating pedestrian traffic patterns along a corridor. Results indicate daily pedestrian traffic volumes can vary significantly between consecutive city blocks, limiting the applicability of a single count location to represent a larger area. Additionally, shifts in temporal traffic patterns occur over short distances. / May 2016

Non-motorized (i.e., bicycle and pedestrian) traffic patterns are an understudied but important part of transportation systems. A key need for transportation planners is traffic monitoring programs similar to motorized traffic. Count campaigns can help estimate mode choice, measure infrastructure performance, track changes in volume, prioritize projects, analyze travel patterns (e.g., annual average daily traffic [AADT] and miles traveled [MT]), and conduct safety analysis (e.g., crash, injury and collision). However, unlike for motorized traffic, non-motorized traffic has not been comprehensively monitored in communities throughout the U.S. and is generally performed in an ad hoc fashion. My thesis explores how to (1) best count bicycles and pedestrians on the entire transportation network, rather than only focus on off-street trail systems or specific transportation corridors and (2) estimate AADT of bicycles and pedestrians in a small college town (i.e., Blacksburg, VA). I used a previously developed count campaign in Blacksburg, VA to collect bicycle and pedestrian counts using existing monitoring technologies (e.g., pneumatic tubes, passive infrared, and RadioBeam). I then summarized those counts to (1) identify seasonal, daily, and hourly patterns of non-motorized traffic and (2) develop scaling factors (analogous to those used in motor vehicle count programs) derived from the continuous reference sites to estimate long-term averages (i.e., AADT) for short-duration count sites. I collected ~40,000 hours of bicycle and pedestrian counts from early September 2014 to January 2016. The count campaign included 4 continuous reference sites (~ full year-2015 counts) and 97 short-duration sites (≥ 1-week counts) that covered different road and trail types (i.e., major road, local road, and off-street trails). I used 25 commercially available counters (i.e., 12 MetroCount MC 5600 Vehicle Classifier System [pneumatic tube counters], 10 Eco-counter 'Pyro' [passive infrared counters], and 3 Chambers RadioBeam Bicycle-People Counter [radiobeam counters]) to conduct the traffic count campaign. Three MetroCount, 4 Eco-counter, and 1 RadioBeam counter were installed at the 4 continuous reference sites; the remaining counters were rotated on a weekly basis at the short-duration count sites. I validated automated counts with field-based manual counts for all counters (210 total hours of validation counts). The validation counts were used to adjust automated counts due to systematic counter errors (e.g., occlusion) by developing correction equations for each type of counter. All automated counters were well correlated with the manual counts (MetroCount R2 [absolute error]: 0.90 [38%]; Eco-counter: 0.97 [24%]; RadioBeam bicycle: 0.92 [19%], RadioBeam pedestrian: 0.92 [22%]). I compared three bicycle-based classification schemes provided by MetroCount (i.e., ARX Cycle, BOCO and Bicycle 15). Based on the validation counts the BOCO (Boulder County, CO) classification scheme (hourly counts) had similar R2 using a polynomial correction equation (0.898) as compared to ARX Cycle (0.895) and Bicycle 15 (0.897). Using a linear fit, the slope was smallest for BOCO (1.26) as compared to ARX Cycle (1.29) and Bicycle 15 (1.31). Therefore, I used the BOCO classification scheme to adjust the automated hourly bicycle counts from MetroCount. To ensure a valid count dataset was used for further analysis, I conducted quality assurance and quality control (QA/QC) protocols to the raw dataset. Overall, the continuous reference sites demonstrated good temporal coverage during the period the counters were deployed (bicycles: 96%; pedestrians: 87%) and for the calendar year-2015 (bicycles: 75%; pedestrians: 87%). For short-duration sites, 98% and 94% of sites had at least 7 days of monitoring for bicycles and pedestrians, respectively; no sites experienced 5 days or less of counts. I analyzed the traffic patterns and estimated AADT for all monitoring sites. I calculated average daily traffic, mode share, weekend to weekday ratio and hourly traffic curves to assess monthly, daily, and hourly patterns of bicycle and pedestrian traffic at the continuous reference sites. I then classified short-duration count sites into factor groups (i.e., commute [28%], recreation [11%], and mixed [61%]). These factor groups are normally used for corresponding continuous reference sites with the same patterns to apply scaling factors. However, due to limitations of the number (n=4) of continuous reference sites, the factor groups were only used as supplemental information in this analysis. To impute missing days at the 4 continuous reference sites to build a full year-2015 (i.e., 365 days) dataset, I built 8 site-specific negative binomial regression models (4 for bicycles and 4 for pedestrians) using temporal and weather variables (i.e., daily max temperature, daily temperature variation compared to the normal 30-year averages [1980-2010], precipitation, wind speed, weekend, and university in session). In general, the goodness-of-fit for the models was better for the bicycle traffic models (validation R2 = ~0.70) as compared to the pedestrian traffic models (validation R2 = ~0.30). The selected variables were correlated with bicycle and pedestrian traffic and cyclists are more sensitive to weather conditions than pedestrians. Adding model-generated estimates of missing days into the existing observed reference site counts allowed for calculating AADT for each continuous reference site (bicycles volumes ranged from 21 to 179; pedestrian volumes ranged from 98 to 4,232). Since a full year-2015 dataset was not available at the short-duration sites, I developed day-of-year scaling factors from the 4 continuous reference sites to apply to the short-duration counts. The scaling factors were used to estimate site-specific AADT for each day of the short-duration count sites (~7 days of counts per location). I explored the spatial relationships among bicycle and pedestrian AADT, road and trail types, and bike facility (i.e., bike lane). The results indicated that bicycle AADT is significantly higher (p < 0.01) on roads with a bike lane (mean: 72) as compared to roads without (mean: 30); bicycle AADT is significantly higher (p < 0.01) on off-street trails (mean: 72) as compared to major roads (mean: 33). Pedestrian AADT is significantly higher (p < 0.01) on local roads (mean: 693) as compared to off-street trails (mean: 111); this finding is likely owing to the fact that most roads on the Virginia Tech campus are classified as local roads. In Chapter 5, I conclude with (1) recommendations for implementation (e.g., counter installation and data analysis), (2) key findings of bicycle and pedestrian traffic analysis in Blacksburg and (3) strengths, limitations, and directions for future research. This research has the potential to influence urban planning; for example, offering guidance on developing routine non-motorized traffic monitoring, estimating bicycle and pedestrian AADT, prioritizing projects and measuring performance. However, this work could be expanded in several ways; for example, deploying more continuous reference sites, exploring ways to monitor or estimate pedestrians where no sidewalks exist and incorporating other spatial variables (e.g., land use variables) to study pedestrian volumes in future research. The overarching goal of my research is to yield guidance for jurisdictions that seek to implement systematic bicycle and pedestrian monitoring campaigns and to help decision making to encourage healthy, safe, and harmonious communities. / Master of Urban and Regional Planning

No description available.

The current state of pedestrian traffic monitoring is characterized by short-duration counts over inconsistent time intervals, making it difficult to compare data temporally at a location or spatially between different locations. Practitioners require understanding of hourly pedestrian traffic periodicities in order to maximize the utility of their short-duration counts. This research deployed six automated pedestrian counters at 12 study sites representing six roadway segments in Winnipeg’s commercial zones. Pedestrian traffic data was collected in 2012 over the summer and fall seasons. This research analyzes the influence of temporal and spatial factors on hourly pedestrian traffic periodicities to enable the characterization of hourly pedestrian traffic in commercial zones. Results indicate that short-duration counts be collected from Tuesday to Thursday on days with less than four hourly precipitation events. Additionally, pedestrian traffic varies seasonally and between adjacent sidewalks in commercial zones. Finally, characterization of pedestrian traffic pattern groups requires detailed land-use data. / October 2015

Honors Project--Smith College, Northampton, Mass., 2010. / Includes bibliographical references (p. 72-74).

Dissertation (PhD)--University of Stellenbosch, 2006. / Bibliography. Also available via the Internet.

We present a novel approach for vehicle detection and tracking in highway surveillance videos. This method incorporates well-studied computer vision and machine learning techniques to form an unsupervised system, where vehicles are automatically "learned" from video sequences. First an enhanced adaptive background mixture model is used to identify positive and negative examples. Then a video-specific classifier is trained with these examples. Both the background model and the trained classifier are used in conjunction to detect vehicles in a frame. Tracking is achieved by a simplified multi-hypotheses approach. An over-complete set of tracks is created considering every observation within a time interval. As needed hypothesized detections are generated to force continuous tracks. Finally, a scoring function is used to separate the valid tracks in the over-complete set. The proposed detection and tracking algorithm is tested in a challenging application; vehicle counting. Our method achieved very accurate results in three traffic surveillance videos that are significantly different in terms of view-point, quality and clutter. / text

The purpose of this research is to analyse pedestrian volumes on multi-use trails in Winnipeg, Canada. The research methodology consisted of collecting continuous automated pedestrian count volumes at seven locations on four multi-use trails in Winnipeg from January 1, 2014 to December 31, 2014. An average pedestrian volume was calculated for each count site over annual, seasonal, and monthly time periods. Pedestrian volumes were found to vary consistently by month of year and hour of day. Day-of-week patterns were not consistent in terms of pedestrian volume. There was a negative relationship between pedestrian volume and rainfall volume and duration, and average daily wind speed. There was a positive non-linear relationship between pedestrian volume and maximum daily temperature. While pedestrian volume correlates with weather factors, variability remains. This suggests that weather analysis may be useful as a complement, but not a replacement of traditional temporal analysis for estimation of pedestrian volumes. / May 2016

Lau Wai Hung. / Thesis (M.Phil.)--Chinese University of Hong Kong, 2006. / Includes bibliographical references (leaves 63-65). / Abstracts in English and Chinese. / abstract --- p.I / 摘要 --- p.II / acknowledgement --- p.III / table of contents --- p.IV / list of figures --- p.VI / Chapter CHAPTER 1 --- introduction --- p.1 / Chapter CHAPTER 2 --- literature review --- p.4 / Chapter 2.1 --- Traffic data collection methods --- p.4 / Chapter 2.2 --- Vision-based traffic monitoring techniques --- p.6 / Chapter 2.2.1 --- Vehicle tracking approaches --- p.7 / Chapter 2.2.2 --- Image processing techniques --- p.10 / Chapter CHAPTER 3 --- methodology --- p.15 / Chapter 3.1 --- Solution Concept --- p.16 / Chapter 3.2 --- System Framework --- p.18 / Chapter 3.2.1 --- Edge Detection Module --- p.20 / Chapter 3.2.2 --- Background Update Module --- p.22 / Chapter 3.2.3 --- Feature Extraction Modules --- p.25 / Chapter CHAPTER 4 --- experiments and evaluation --- p.41 / Chapter 4.1 --- Setup and Data Collection --- p.41 / Chapter 4.2 --- Evaluation Criteria --- p.42 / Chapter 4.3 --- Experimental Results --- p.44 / Chapter 4.3.1 --- Comparing overall accuracies --- p.44 / Chapter 4.3.2 --- Accuracies for different traffic conditions --- p.46 / Chapter 4.3.3 --- Comparing balanced sampling and random sampling --- p.48 / Chapter 4.3.4 --- Comparing day and night conditions --- p.50 / Chapter 4.3.5 --- Testing on time-series of images --- p.52 / Chapter CHAPTER 5 --- analysis --- p.54 / Chapter 5.1 --- Strengths and Weaknesses --- p.54 / Chapter 5.1.1 --- Sobel Edge Histogram --- p.54 / Chapter 5.1.2 --- Horizontal Line Detection --- p.55 / Chapter 5.1.3 --- Block Detection --- p.56 / Chapter 5.1.4 --- Combined Learning --- p.57 / Chapter 5.1.5 --- Overall Framework --- p.58 / Chapter 5.2 --- Future Research --- p.59 / Chapter 5.2.1 --- Static image based monitoring combined with other traffic monitoring approaches --- p.59 / Chapter 5.2.2 --- Horizontal Line Detection as tracked features of vehicles --- p.60 / Chapter 5.2.3 --- Application in aerial image-based system --- p.60 / Chapter CHAPTER 6 --- conclusion --- p.62 / bibliography --- p.63 / appendix a sobel edge detection --- p.66 / appendix b neural network setup --- p.67 / appendix c numerical results --- p.69

Thesis (M.S.)--University of Texas at El Paso, 2008. / Title from title screen. Vita. CD-ROM. Includes bibliographical references. Also available online.