Detection of the Passenger Density on Public Transport Vehicles in Antalya Using Image Processing Techniques
Detection of the Passenger Density on Public Transport Vehicles in Antalya Using Image Processing Techniques
Muhammed Emin ÖZGÜNSÜR
Department Of Remote Sensing and Geographic Information Systems, Akdeniz University
Abstract — In Antalya which is a tourism city, has the variable population and economic structure. The rapid increase in the intensity of urban traffic and the lack of steps to be taken in the past have raised the need for people to use public transport. However, it is not preferred unless public transportation vehicles are compulsory for reasons such as the inadequate capacity of the public transportation, inaccessibility of time, cost of transportation, unknowable behavior of the drivers.
It is essential to know the locations where settlement places, road densities, road works, the number of public transportation vehicles, stops and densities according to specific days and hours, public transportation such as hospitals and schools should be easily accessible in order to realize a healthy route and rotation plan. In this research for the obtain data, passenger density detection of public transportation vehicles with camera images used.
Keywords — image processing, public transportation, Antalya
Antalya, a tourism city, is one of the oldest settlements known in Turkey which is based on B.C. 220 thousand years ago. The population is 1,978,333, while 1,392,974 people live in the city center and 585,359 people live in the borders. So, about 70% of the population lives in Antalya city center. In terms of migration, Antalya is the second most migrating city in Turkey with a rate of 8.4% . In Antalya, where 19 districts and 910 localities are located, a household spends an average of 18.8% of its income. The number of domestic tourists coming to Antalya, which hosts extraordinary activities such as the G-20, Expo 2016, is 1,223,795 while the number of airplanes coming from abroad in 2013 is 10.825.977 .
Rapid population growth, increase in migration from the village to the city and the number of tourists causes the demand for public transportation to increase day by day and bring about the rescheduling of the system. In Turkey, the management and control of public transportation belong to the municipalities. Local governments’ efforts to create employment sources in public transportation policies cause significant disadvantages in the system. Because of the large number of vehicles they serve, the difficulties in auditing and the selfish behaviors that drivers show for in order to transport more passengers cause traffic safety to decrease, increase congestion and increase traffic accidents .
Urban public transportation types can be classified under three main headings based on capacities :
- Low Capacity Types: Intermediate Transport
– Calling a vehicle by phone
- Medium Capacity Types : Superficial Transport
– Trolley Bus
- High Capacity Types: Fast Transport
– Fast bus systems
– Light rail systems
– Suburban trains
Comparisons in the literature show that for any given criterion, public transportation types have very different boundary values, or even conflicting situations. For example, in terms of hourly passenger capacity, some sources show light rail systems ahead of fast bus operation, while in some sources this is the opposite. This difference is sometimes made deliberate, in other words, to be brought forward in a certain way. Sometimes this variation is due to the performances of the samples representing the species. In particular, it is not easy to distinguish light rail systems from subway systems, and this causes differentiation in evaluations. Hourly passenger transportation capacities of public transportation vehicles vary . The general approach, however, is that passenger transport capacity:
– Bus, trolleybus, tram systems capacities 7.000 – 15.000 passenger/hour/direction
– Fast bus systems and light rail systems capacities 15.000 – 25.000 passenger/hour/direction
– Subway and suburban train systems 40.000 – 60.000 passenger/hour/direction
II. DEFAULT SYSTEM MODEL
Public transportation vehicles used in Antalya public transportation system (PTS), sea buses and minibusses in district transportations, buses, light rail trains and trams are used by city center transportation. Buses are the busiest public transportation vehicles that experience the most densities which have about 1207 vehicles and about 84 routes in Antalya public transportation system. The number of daily carried citizens are about 400.000 people.
Figure 1: Daily carried citizens with Antalya public transportation system
Today, in Antalya public transportation rotation plan, each rotation has about 15 schedules and each scheduled bus have about 8:00 am. So each route has about 120 voyages. For the 84 routes, total voyages are about 10.080 daily. Even at 10.080 daily voyages, citizens complain that public transportation is not coming in time and that it is full. The rotation plans are daily made. Also not included a number of passengers carried each time for rotation planning. Generally, in Antalya, a single route is not enough to move from one point to another and every day hundreds of people use more than one vehicle.
Table. 1: Daily transfer counts for sample route KL08 to another route
It is necessary to know the locations where settlement places, road densities, road works, the number of public transport vehicles, stops and densities based on specific days and hours, public transportation such as hospitals and schools should be easily accessible in order to realize a healthy route and rotation plan.
In the current public transport system in Antalya, citizens complain about they do not arrive on time and that public transportation vehicles do not arrive on time. If we give priority to citizens’ complaints, the most important of all these information in the creation of the rotation plan is the number of vehicles available and the passenger density at the stops at certain days and hours. With this information, a comfortable and punctual public transportation system can be created with a quick and high-quality public transportation system.
For these reasons, it is essential to determine the passenger density in the vehicle based on the stops in order to make a healthy route and rotation plan.
In Antalya public transportation system, while the number of passengers can be obtained from the stops via the card system, the number of passengers descending at the stops cannot be determined.
For the security solutions, computer vision is implemented entire world ranging. Recent years researchers showed that the automatic passengers counting systems (APCS) with video-based popularity are increasing. Avoid “empty routes” and reduce environmental pollution allows a rational schedule of transportation based on the passenger flow. For the complicated task of passenger counting is bus passengers physical dimensions, outfit, differ in their look.
Also the shadows, solar positions have a lot of influence on signal quality for backgrounds .
Generally, regarding human and object detection have been proposed different feature descriptors. People have detected their head-shoulder parts based on the omega-shape features. AdaBoost and Viola-Jones classifications use the histogram of oriented gradients features are combined to obtain reliable and fast results .
For the passenger density detection with camera generally used public transportation vehicles doors and detect in/out movement. One of the research for Kaunas public transportation buses to collecting real-life video data information with passenger density detection used door cameras. The results of research-tested 32 in and 38 out situations for the single passenger. The success of the detection is 86% accuracy for one passenger passed each time. But the multiple passengers passing each time to accuracy is 47% .
Generally, the door switching detection is a prerequisite for accurate passenger counting because, during the process of the bus moving, many environmental disturbances will influence the performance of the vision-based counting system .
To detect the passenger density in public transportation vehicles, sensors can be installed in the door entrances, the system can be placed in the system where the weight will be measured, and passengers can be counted by image processing. And it is aimed to create data for rotation planning by determining the passenger density by image processing technique using images taken from low-resolution security cameras in vehicles .
III. DETECTION OF THE PASSENGER DENSITY
The detection of passenger density in public transportation vehicles used vehicles security cameras which generate 352 x 288-pixel low-resolution camera images. Since these pictures are in low resolution, it is very misleading to determine the density with the number of pictures in the picture [Fig. 2]. When these camera images are examined, it has been determined that even human eyes can not be seen and counted by people in public transportation, especially in the rear.
It has not been considered to increase the number of cameras on the camera to set high-resolution cameras in order to try to determine the passenger density without additional cost with the existing system. For these reasons, another method with image processing has been preferred for determining the passenger density.
Figure 2: Detected humans from vehicle cameras with image processing
The second preferred way to determine the density is to take the idle view of the public transportation vehicle and optimize it to remove it from the other vehicle images by image processing. Although the number of total passengers cannot be determined on this research, it would be possible that bus density will be known approximately.
The detected image difference will give an approximate density value and these density values will be classified as empty, less dense, dense, over-dense. Through these transactions, it is thought that it is possible to determine which routes, which vehicles, at which times and on which dates they are used heavily. With more detailed analyzes, it can be determined which stops are busy and which stops after which stops. With these determinations made, the empty public transportation lines can be rearranged and existing public transportation vehicles can be used more efficiently.
Figure 3: Empty and less dense density bus image for bus 01
In the empty and less dense bus images obtained in Figure-3, the images are converted to black-and-white images so that the image difference can be taken properly, and then the differences are taken and filtered. As a result of the work done, white areas give passenger areas.
Generally use three ways for the voting foreground extraction which optical flow, frame-to-frame difference and background modeling methods. One of this optical flow method is unfavorable and complex for real-time processing. The other method is the frame-to-frame difference which more dependent on the speed of passengers getting on and off. Both methods have limitations and select background modeling method because the installed scene is fixed. A high accuracy and robustness Hybrid Gaussian model interference can be eliminated by morphological processing and the non-foreground area could eliminate by area filtering which connected component detection is down. After the area of each connected component is calculated, the area is smaller than a thresh, the pixels of the connected component are all set to zero. The extracted foreground contains mainly the passengers themselves, with the majority of the heads, arms, and shoulders .
After capturing the foreground image, locate the potential head region in the foreground image. The head is obviously a middle high and low side shape. Using this feature, located area of the head as followed like divide the image as several patches equally, calculate the average height of each patch and if the height of a patch is higher surround the patches which is possible a part of the head and selected it as candidate patch .
Figure 4: Result of image processing bus images for bus 01
The ratio of the white area to the black area gives the passenger density. According to this ratio, the passenger density is 29.46% compared to the pixel ratio, which can be classified as less intensive [Fig. 4].
Figure 5: Empty and over dense density bus image for bus 02
Another example, Figure-5, shows empty and over dense bus security camera images. In this view, the light reflections are seen different from Figure-3.
Figure 6: Result of image processing bus images for bus 02 with reflection
When image processing is performed without regard to light reflection, the pixel-based passenger density ratio is 31.54%. Better results were obtained in the case of higher density ratio than in Fig. 3 with low density and in the case of light reflection removal, despite the end result.
Handling occlusion of the track people used image long-memory and segmentation. The thresholding the result to highlight these objects of interests are background subtraction to achieve areas of interest. Extraction on the resulting blobs, areas, perimeter, bounding box, height, width, and the centroid is the systems track them from frame to frame and keep the track of all the tracking information. The storing tracking data for each blob, system manage occlusion fairly well .
In the current public transportation system in Antalya, citizens complain that they do not arrive on time and that public transportation vehicles do not arrive on time. It is mandatory to determine the passenger density in the vehicle based on the stops in order to make a healthy route and rotation plan. The detection of passenger density in public transportation vehicles used vehicles security cameras which generate 352 x 288-pixel low-resolution camera images. For the image processing has been preferred for determining the passenger density.
The detected image difference will give an approximate density value and these density values will be classified as empty, less dense, dense, over-dense. In the empty and less dense bus images, the images are converted to black-and-white images so that the image difference can be taken properly, and then the differences are taken and filtered. As a result of the work done, white areas give passenger areas. With these determinations made, the empty public transportation lines can be rearranged and existing public transportation vehicles can be used more efficiently.
 Türkiye İstatistik Kurumu Verileri, Bölgesel Göstergeler TR61 Antalya, Isparta, Burdur, 2010 Ankara
 Türkiye İstatistik Kurumu Verileri, Seçilmiş Göstergelerle Antalya, 2013 Ankara
 Hüseyin Ceylan, Özgür Başkan, Soner Haldenbilen, Halim Ceylan, Şehir İçi Toplu Taşım Sorunları ve Çözüm Yöntemleri Denizli Örneğin, 5. Kentsel Altyapı Ulusal Sempozyumu Bildirisi
 Mustafa Tanış, Kemal Selçuk Öğüt, Orta Ölçekli Kentler İçin Toplu Taşıma Seçeneklerinin Teknik ve Mali Karşılaştırılması, İstanbul Teknik Üniversitesi
 Feng Li, Fuwei Yang, Haiwei Liang, Wenming Yang, Automatic Passenger Counting System For Bus Based On RGB-D video, Advances in Engineering Research (AER) volume 17
 P. Lengvenis, R. Simutis, V. Vaitkus, R. Maskeliunas, Application Of Computer Vision Systems For Passenger Counting in Public Transport, Elektronika IR Elektrotechnika volume 19
 P. Potter, P. Belet, C. Poppe, S. Verstockt, P. Lambert, R. Walle, Passenger Counting In Public Rail Transport Using Head-Shoulder Contour Tracking, Proceedings Of The 7th International Conference On Computer Vision Theory and Application 705-708
 Ph.D. Daniel Moura, Eng. Eduardo Marques, Image Descriptors For Counting People With Uncalibrated Cameras, Mestrado Integrado Em Engenharia Eletrotecnica E De Computadores July 30, 2014
 N. V. Rajesh Kumar, G. Dhanasekar, M. Dennis, Application Of Face Detection System For Passenger Counting In Lifts Using Haar Features, ARPN Journal Of Engineering And Applied Sciences volume 11
 Tao Yang, Yanning Zhang, Dapei Shao, Ying Li, Clustering Method For Counting Passengers Getting In A Bus With Single Camera, Optical Engineering volume 49