A method of measuring of a people flow along a city street and a device therefor

Abstract

A method of measuring of a people flow along a city street using a series of photos taken at time interval, according to the invention comprises a step of acquiring (210) street network topology comprising base segments and respective calibration coefficients and a step of acquiring a series of photos (221) of the street taken from at least one moving vehicle and recording time and location of taking each photo and a digital image representing photo. Then the method comprises a step of detecting (222) people in the digital images to obtain candidates, a step of assigning (223) relative positions to the candidates. The method comprises a step of snapping (231) the locations to the street network topology and a step of determining a trajectory (232) of the vehicle with respect to the street network topology. The obtained data are used in a step of filtering a set of candidates (233) to obtain only the candidates on the side closer to the vehicle trajectory, a step of filtering out (234) the candidates having coordinates outside the base segment. Remaining candidates are counted (235) and result is corrected (236) according to calibration coefficients and finally a step of outputting (273) obtained flow of people. A mobile device according to the invention has a camera, a clocking device and a positioning system receiver adapted to realize the method according to the invention. A computer program product according to the invention is adapted to realize the method according to the invention.

Description

A method of measuring of a people flow along a city street and a device thereforFIELD OF THE INVENTION

[0001] The invention concerns a method of measuring of a people flow along a city street .STATE OF THE ART

[0002] European patent application EP3425605 discloses a people flow evaluation system comprising : a receiving unit for receiving people flow information, an input unit for inputting people flow control information, a data management unit for storing people flow control information data, a control function operation unit for calculating control functions from the people flow control information and the people flow control information data, a people flow operation unit for performing simulation processing on people flows from the people flow information and the control functions ; and an ef fect operation unit for calculating a people flow control ef fect on the basis of the people flow simulation results of the people flow operation unit .

[0003] CN110717926A discloses a method and device for obtaining pedestrian flow information by processing of images from stationary devices . The method comprises : acquiring the position of each entrance and exit in M entrances and exits in a target area according to a first track set , wherein the track set comprises the motion track of at least one pedestrian passing through the target area . According to the position of each passageway, a second track set of each passageway in N passageways is determined . N passageways are determined according to the M passageways , and the second track set of the passageways comprises movement tracks of pedestrians walking in the passageways in the first track set . Pedestrian flow information is acquired of each passageway according to the second track set of each passageway .

[0004] CN112287778 discloses a method of pedestrian volume analysis based on direction aggregation . Method comprises stepsof establishing a coordinate system, synthesi zing an obtained target picture on the coordinate system so as to generate person tracks and fit discrete points on each person track to form a straight line ; then dividing the synthesi zed target picture into a plurality of dynamic regions , combining the straight lines with similar directions in the dynamic regions for multiple times , and finally generating a pedestrian volume analysis interface . According to the method, traj ectory curves of sorted figures are fitted to form straight lines , dynamic areas are scienti fically and reasonably divided, the straight lines with similar directions in the dynamic areas are merged for multiple times , the pedestrian flow direction of the whole area is sorted and analyzed and provided for a user as a reference . Through statistics and analysis of pedestrian flow data, the resources are utili zed better .

[0005] Publication W02017159060 of international patent application discloses an information processing device includes a first analysis unit , a second analysis unit , and an estimation unit . The first analysis unit calculates the flow of a crowd in the image pickup range of a fixed camera using a first surveillance image . The second analysis unit calculates the distribution of the attributes of an obj ect in the image pickup range of a moving camera using a second surveillance image . The estimation unit estimates an attribute distribution for a range not included in the image pickup range of the moving camera .PROBLEM TO BE SOLVED

[0006] Techniques known in the art rely predominantly on stationary imaging devices . Therefore analysis is limited to places in which imaging devices are available . I t is an aim of the present invention to provide a method that allows uni form methodology applicable to determine people flow on large areas , of cities not only stationary points allowing applying the same rules of counting in the whole area .SUMMARY OF THE INVENTION

[0007] The invention concerns a method of measuring of a people flow along a city street using a series of photos taken at time interval . The method comprises a step of acquiring street network topology comprising base segments and respective calibration coef ficients and a step of acquiring a series of photos of the street taken from at least one moving vehicle and recording time and location of taking each photo and a digital image representing photo . Then the method comprises a step of detecting people in the digital images to obtain candidates . Then relative positions are assigned to the candidates . Photos and observer are snapped to the street network topology . A step of determining a traj ectory of the vehicle with respect to the street network topology provides information about coordinates and orientation of the vehicle . Obtained information is used in a step of filtering a set of the candidates to the candidates on the side closer to the vehicle traj ectory . A step of filtering out the candidates having coordinates outside the base segment precedes a step of counting the remaining candidates in the base segments . Results of counting the candidates in base segments are corrected according to calibration coef ficients and obtained flow of people is outputted .

[0008] The invention concerns a mobile device having a camera, a clocking device and a positioning system receiver adapted to reali ze the method according to the invention .

[0009] Further the invention concerns a computer program product adapted to reali ze the method according to the invention .BRIEF DESCRIPTION OF FIGURES

[0010] The invention has been described in details with reference to attached drawings .

[0011] Fig . 1 shows a flow chart of the method according to the invention together with preceding calibration steps .

[0012] Fig . 2 shows a flow chart of the method according to the invention using previously acquired calibration coef ficients and street network topology .DESCRIPTION OF EMBODIMENTS OF THE INVENTION

[0013] The method according to the invention to be success fully run requires preceding calibration of the street network and the mathematical model .

[0014] Fi rst the street network needs to be acquired in a step of street network acquisition 111 . Acquisition may be a mere loading of the digital map or digital processing of a satellite photo .

[0015] Over the acquired street network a step of segmentation 112 is run to obtain base segments . In the present example length of base segment is 30 m . Generally it is reasonable to select base segments in range from 10 to 50 m .

[0016] Simultaneously in acquisition step 121 photos are taken by an observer from a vehicle moving a long a city street . Observer is a logical construct as photos are taken automatically . The observer has own coordinates and orientation with respect to the map - or more precisely the street network topology . Photos are taken at time intervals , and each photo is stored together with position at which it has been taken .

[0017] Subsequently people are recogni zed at photos in a step of recognition 122 . In this step people are recogni zed via processing the photo by neural network to obtain "candidates" . Then - in a step 123 of assigning candidate position - every "candidate" is assigned relative position with respect to the observer - a position with respect to camera that has taken a picture . Relative position is represented as an angle and a distance .

[0018] The street network topology and a present position corresponding to the photo are used to snap the observer coordinates to the street in a step of snapping 131 . This operation results in corrected position of the observer foranalyzed photo . The photo and a sequence of positions are in the same step processed so as to determine traj ectory of the observer and an orientation of the observer with respect to the street network topology . Position of the observer and orientation of the observer are stored as observer coordinates which are result of the step of snapping 131 .

[0019] The observer coordinates and relative coordinates of candidates are used to enhance assigning people to the segments in the street network topology . In the next step of side filtering 132 candidates form the opposite side of the street with respect to the observer are filtered out . Only the candidates located in the pavement on the closer side of the street network topology segment are taken into further consideration . In most of the countries the closer side is the right side of the observer . The candidates outside analyzed segment of the street network are filtered out in the step 133.

[0020] After processing the photos along the observer traj ectory, at this step the street network topology comprising base segments is known and number of people in the base segments are known . This is enough to start estimating people a volume of people (pedestrian) flow in each base segment .

[0021] In the present embodiment a model based on the average speed of pedestrian movement in the city was adopted . This speed usually is within a range between 1 and 1 . 6 m / s . In the present example 1 m / s was assumed .

[0022] Accordingly a pedestrian captured in base segment spends in the segment 30 s in average . When a person is detected every 30 s in the segment , then it can be assumed that it is 2 persons per minute and 120 persons per hour . This can be illustrated by the formula : n • v N = 3600 - s where N denotes a number of pedestrian per hour in the segment , n denotes a number of pedestrians detected in the segment and sis a length of the segment. Thus detected pedestrians are counted in the step of counting 134.

[0023] To get a real value a number N needs to be adjusted according to a coefficient "a" representing effectiveness of detection. It is typically around a — 0.5 and represents a fraction of pedestrians detected with respect actual number. This coefficient is different than "1" because of shielding by other vehicles, infrastructure or city plants. This can be estimated from the photo or even assigned to the street network topology base segment as a parameter. It is determinable thanks to the limitation to the same side of the road. Attempts to find reliable coefficients for the pedestrians on the opposite sided proved to be highly ineffective.

[0024] Finally the corrected number is represented as:

[0025] Coefficient "A" depends on the effectiveness of detection and average pedestrian velocity in the base segment. "A" can be determined experimentally in calibration by comparing results with other measurements e.g. with stationary measurements. Stationary measurements provide ground truth in step 141 which is then processed in the step 142 into reference people flow data. Then the coefficient "A" can be determined by comparison of the stationary data with data taken from mobile vehicle for the same base segment.

[0026] Exemplary tests in Warsaw (Poland) for base segments located at street address Marszalkowska 24 and Malczewskiego 6 allowed to obtain A=301,7 for the former and A=180,8 and analysis should be performed in buffers of 3 segments has been corrected to 100 m.

[0027] During the calibration step 150 coefficients "A" are computed and assigned to the base segments in the neighborhood of the stationary measurement. Length of the base segments may also be reconfigured for different parts of the street networktopology so as to minimi ze a deviation from stationary measurements in the tests or analysis can be done in buf fers of multiple segments . Therefore the calibration results in calibration coef ficients and corrected street network topology returned in the step 151 .

[0028] Once calibration is completed the method can be reali zed using corrected street network topology and calibration coef ficients . A flow chart of the method is shown in Fig . 2 . Accordingly the method of measuring of a people flow along comprises a step of acquiring 210 street network topology prepared previously comprising base segments and acquiring respective calibration coef ficients . Then a step of acquiring a series of photos 221 of the street taken from at least one moving vehicle is executed . For each photo time and location of taking each photo and a digital image representing photo are recorded . The photos can be taken e . g . with smartphone used by a taxi driver . Although said mobile device can be used to acquire photos and process them by executing method according to the invention to determine people flow it is better to use more cars and more cameras and upload the photos and coordinates to the server processing them all . In that scenario the method according to the invention is executed by the server or even better a server swarm e . g . Docker Swarm . When multiple vehicles are used it is better to take photos in the series in random times to avoid correlation with regular daily events .

[0029] Subsequently a step detecting 222 people in the digital images to obtain candidates is executed . Candidates are then assigned (223 ) relative positions with respect to the camera ( observer ) taking the photo . Positions are represented by angle and distance .

[0030] Subsequently a step of snapping 231 the locations to the street network topology and a step of determining a traj ectory 232 of the vehicle are executed . The snapping consist in casting the coordinates from the positioning system to the street networktopology . It is done with taking into account the previous position of the vehicle and the direction of movement . The traj ectory is determined from the series of subsequent coordinates . It includes determining orientation of the camera with respect to the street network topology .

[0031] Traj ectory can be corrected iteratively taking into account velocity of the vehicle as determined with positioning system . The positioning system preferably is a global positioning system ( GPS ) . Velocity in subsequent positions are used to determine acceleration which is used in the next iteration . Preferably an imaging device - a camera or smartphone - is provided with acceleration sensor which is stored . Then the acceleration data computed with GPS are compared with data from acceleration sensor and correction is applied to iteratively reduce the discrepancy .

[0032] The traj ectory data, orientation data are used to execute a step of filtering the set of candidates 233 to candidates on the side of the street closer to the vehicle traj ectory .

[0033] Then additional step of filtering out 234 the candidates having coordinates outside the current base segment is executed . Accordingly only candidates within each base segments are then counted in the step of counting 235 . The number of counted candidates is then corrected according to calibration coef ficient in the respective base segments . Finally resulting people flow is outputted 273.

[0034] People flow data can be used for regulating street lights , planning reconfigurations of pavements and analyzing safety . The invention allows taking the measurements of people flow without a need for stationary infrastructure using available mobile vehicles in much lower numbers . The invention of fers uni form, method of assessment and any di f ferences with respect to quality of used hardware e . g . imaging devices are not correlated to location as they concern mobile vehicle .

Claims

CLAIMS1 . A method of measuring of a people flow along a city street using a series of photos taken at time interval , characteri zed in that it comprises a step of acquiring (210 ) street network topology comprising base segments and respective calibration coef ficients , a step of acquiring a series of photos (221 ) of the street taken from at least one moving vehicle and recording time and location of taking each photo and a digital image representing photo , a step of detecting (222 ) people in the digital images to obtain candidates , a step of assigning (223 ) relative positions to the candidates , a step of snapping (231 ) the locations to the street network topology, a step of determining a traj ectory (232 ) of the vehicle with respect to the street network topology, a step of filtering a set of candidates (233 ) to obtain only the candidates on the side closer to the vehicle traj ectory, a step of filtering out (234 ) the candidates having coordinates outside the base segment , a step of counting (235 ) the remaining candidates in the base segments , a step of correcting (236 ) results of counting the candidates in base segments according to calibration coef ficients , a step of outputting (273 ) obtained flow of people .2 . A mobile device having a camera, a clocking device and a positioning system receiver adapted to reali ze the method of claim 1 .3 . A computer program product adapted to reali ze the method of claim 1 .

Images