Using existing surveillance infrastructure to monitor pedestrians on pedestrian bridges: a proof of concept

Date
2016-12
Journal Title
Journal ISSN
Volume Title
Publisher
Stellenbosch : Stellenbosch University
Abstract
ENGLISH ABSTRACT: Due to the high number of pedestrians involved in crashes along the freeways in South Africa, there is a need for authorities to monitor pedestrians using the freeways in order to gain valuable mobility information. As South Africa is a developing country, existing surveillance infrastructuremust be used to its fullest extent. This work presents a proof of concept that existing surveillance infrastructure can be used to implement a pedestrian monitoring system. The system willmonitor pedestrians using pedestrian bridges along the freeways, and will collect pedestrian mobility information, such as the direction and relative speed of movement. The system thus aims to contribute to pedestrian safety in South Africa. A cascade classifier is trained with surveillance footage collected from SANRAL using AdaBoost and the use of approximated image scales. It is shown that the detectors need to be trained specifically for a scene/camera using footage from that scene/camera in order to compete with modern pedestrian detection applications. The Integral Channel Features (ICF) detection method, which incorporated HOG and colour channels, is used in this thesis as it is easily implemented and achieved stateof-the-art performance. A graphical user interface is created to run the training and detection phases as this removes the user formthe complicated back-end. The detected pedestrians are tracked using parametric tracking methods such as a Kalman filter, and supplementary track management functions. The tracked pedestrians are automatically counted and compared to pedestrian counts done by hand. The results show that the automatic pedestrian counting system achieved an accuracy of more than 90%. A user friendly verification user interface is created to allow for themeaningful representation the pedestrian mobility results. A further analysis of the pedestrian mobility information collected, shows that the peak hour pedestrian traffic in the morning is between 06:30-07:30, and in the afternoon between 17:00-18:00. The data also shows that pedestrian traffic in the morning is highest towards the CDB, and in the afternoon is away from the CBD. It can also be seen that pedestrians move on average 10% faster towards the CBD. The results is proof that the pedestrian monitoring system using existing lowresolution surveillance infrastructure on pedestrian bridges can be used to obtain state-of-the-art performance. Thus it demonstrates that the system achieves the goals set, and that the application of such a system is to be kept in mind during future infrastructure installations and upgrades.
AFRIKAANSE OPSOMMING: As gevolg van die hoë aantal voetgangers betrokke in ongelukke langs die snelweë in Suid Afrika, is dit noodsaaklik vir die betrokke owerhede om voetgangers te monitor en so belangrike bewegingsinligting te genereer. Suid Afrika ’n ontwikkelende land en dus moet die bestaande veiligheidsinfrastruktuur ten volle benut word. Hierdie tesis dien as ’n bewys dat bestaande infrastruktuur gebruik kan word vir die implementering van ’n voetganger-waarneemingsisteem. Die stelsel sal voetgangers monitor wat die voetganger brue langs die snelweë gebruik, en sal bewegingsinformasie soos, rigting en relatiewe spoed van beweging, genereer. Die stelsel beoog dus om by te dra tot die veiligheid in Suid Afrika. ’n Kaskade klasifiseerder is opgelei met bestaande veiligheidsbeeldmateriaal wat van SANRAL verkry is, deur gebruik te maak van AdaBoost en geskatte beeld-skale. Daar word aangetoon dat ’n klasifiseerder spesifiek opgelei moet word vir die kamera wat gebruik gaan word. Die Integral Channel Features klasifiseerdingsmetode is gebruik, wat HOG en kleur kanale inkorporeer, omdat dit maklik implimenteerbaar is, en goeie resultate behaal. ’n Grafiese gebruikerskoppelvlak is ontwikkel om die opleidings- en klassifiseringsfases uit te voer sodat die gebruiker nie bloot te stel is aan die ingewikkelde program-agterkant nie. Die waargenome voetgangers word gevolg deur parametriese volgmetodes soos ’n Kalman filter en aanvullende spoor-besturendsfunksies. Die gevolgde voetgangers is automaties getel en vergelyk met voetgangers wat met die hand getel is. Die resultate dui daarop dat die automatiese voetganger telling ’n akkuraatheid van meer as 90% behaal. ’n Gebruikers vriendelike grafiese koppelvlak is onwikkel om voorsiening te maak vir betekenisvolle voetganger bewegings resultate. Verdere ontleding van die voetganger bewegings resultate wat ingesamel is, toon dat die spitstyd voetgangerverkeer in die oggend tussen 06:30-07:30 is, en in die middag tussen 17:00-18:00. Die data toon ook dat die meeste voetgangersverkeer soggens in die rigting van die middestad is, en in die middag weg van die middestad is. Die resultate toon ook dat voetgangers gemiddeld 10% vinniger beweeg in die rigting van die middestad. Die resultate dien dus as bewys dat ’n voetganger moniteringstelsel wat gebruik maak van bestaande lae-resolusie veiligheids infrastruktuur op voetganger brue, gebruik kan word om kompeterende resultate te behaal. Dit demonstreer dat die stelsel die doelwitte behaal wat gestel was, en dat die implimentering van ’n voetganger waarnemingsstelsel iets is wat ingedagte gehoumoet word tydens toekomstige infrastruktuur installasies en opgraderings.
Description
Thesis (MEng)--Stellenbosch University, 2016.
Keywords
Pedestrian areas -- Design and construction, Elevated walkways, Footbridges, UCTD
Citation