The Hong Kong 7/1 March: Crowd Size Estimates (Ming Pao Aerial Photo Analysis)

SOURCE #3:  MING PAO AERIAL PHOTO ANALYSIS says 192,000

    

(Ming Pao)  Computer analyses showed 264,000 marchers

[Translated summary]  Was it 200,000?  530,000?  Every time there is a large-scale event, the number of marchers is a disputed figure.  We asked a satellite photo analysis specialist to examine the photos that we took from a helicopter.  The final estimate is 264,000.  

According to our expert Thomas Lee, in the Victoria Park soccer fields, there were about 100 persons for every 100 square meters.  Based upon a road surface of 72,000 square meters (=3,600 meters long and 20 meters wide), there were about 72,000 persons at 4pm.  Under the assumption of a travel time of 90 minutes, there should be a total of 3.67 groups in the 5.5 hour time span.  Therefore, the total estimated crowd size is 3.67 x 72,000 = 264,000 persons ignoring those who joined in the middle.  The margin of error is 15%.

According to Civil Human Rights Front spokesperson Jackie Hung, the police estimated that the route capacity is 170,000 persons whereas the six soccer fields in Victoria Park contains 110,000 persons.  Hung does not believe that the route can hold only 70,000 persons, because there has to be more on the road than inside Victoria Park.

20萬﹖53萬﹖每逢大型遊行活動,遊行人數多少必是爭議所在。本報昨日委託衛星圖片分析專家,嘗試利用太空科技的衛星圖片分析軟件,以科學化的方式,將本報從高空拍攝的相片作出分析,務求探討真正的「遊行人數」,最後推算出昨日參與的遊行人數為26.4萬。

從事衛星圖片科技研究,身兼香港大學地理系遙感課程(remote sensing)及理工大學衛星定位(GPS)應用課程的客席講師的星眺有限公司負責人李偉鵬,昨日透過一種名為「太空科技衛星圖片分析軟件」(image remote sensing software),用多光譜分析原理,配合人體的大小及形狀,從空中圖片中找出「一個個的人型物體」作評估,然後統計特定空間內的人群數量。

本報記者於昨午4時半,從直升機以接近垂直的角度,於高空拍攝維園的人群聚集情,並將圖片以軟件分析出100平方米內的人數,並透過準確的數碼地圖,計算出遊行路段總面積於一個時間的總人數。

分析顯示,由維園至政府總部遊行指定路段,全長3600米,而街道的闊度則不一,如高士威道闊24米、軒尼詩道闊24米,但要扣除當中作緊急行車的4米電車路,故整段道路平均闊20米作計算。

李偉鵬指出,軟件分析顯示每維園球場中,每100平方米約有100人,而最疏的人群則為85人,如以100平方米有100人,全路段面積約7.2萬平方米(3600×20米)推算,在昨午4時半的遊行路面上約有7.2萬人。按90分鐘一程人群,5.5小時內共有3.67程,故遊行人數推算為26.4萬人(未計中途插隊人數)。而軟件偏差率為15%

不過,民間人權陣線發言人孔令瑜指出,港島總區行動部高級警司梁文幹曾指,整段遊行路線(封鎖共6條行車線時)可容納17萬人,而維園6個足球場加上草地場則可容納11萬人。

孔令瑜強調,民陣計算53萬人參加遊行是在這個基礎下計算出來的,因為整條路需走90分鐘,遊行人士共走了近5小時、即3程多,計算出53萬人是絕對有科學根據的。對於衛星圖片顯示整段遊行路程只可容納約7萬人,孔令瑜並不相信,認為該段路容納人數不可能比維園還少。

其實,衛星圖片分析軟件一直於本港應用,包括用於分析本港水質、土地利用、植物如紅樹林的分佈,李偉鵬表示,去月曾替發展商,為內地評估一個荔枝園的大細,以便作收地評估之用,而今次也是首次利用此軟件分析人群。

This project in fact consisted of two phases.  In the first phase, Ming Pao hired a helicopter and took aerial photos.  These photos were then submitted to the satellite photo analysis expert Thomas Lee, who used his specialized software to count the number of people.  Thomas Lee is a honorary and visiting lecturer at Hong Kong University (Remote Sensing and Photogrammetry) and Hong Kong Polytechnic University (Global Positioning System in Logistics), where he teaches classes on how to use remote sensing software to count tree density, identify land use, position objects, etc.

Although many photos were taken, not all were suitable for use.  For example, it would be difficult to discern people after it got dark; or when the buildings cast deep shadows on the marchers.  So the best photos were those that were taken while the marchers assembled on the open soccer fields in Victoria Park.  On the basis of those photos, Thomas Lee estimated that the average density was 100 persons per 100 square meters.  He also noted that he counted sometimes as few as 85 persons per 100 square meters in some photos but other times as many as 115 persons per 100 square meters elsewhere.  This is not a statement of the margin of errors as such, but the natural variation in the data distribution itself.  Thomas Lee actually recommended that the less dense factor of 85 per 100 square meters ought to be use.  

Thomas Lee also used accessed a digital map database and geographical information system (GIS) to study the march route.  He determined that the total length of the route was 3,600 meters.  The width of the road was 20 meters on averages, although it could be more or less in some places.  In the end, he used the 20 meter figure to obtain an estimate of the total area being 20m x 3,600m = 72,000 square meters.  In addition, he also recommended adjustments to be made for people leaving before the end or joining after the start. 

This was the information that Thomas Lee provided to Ming Pao.

In the second phase, Ming Pao took that information and made their own estimate of the march size.  In so doing, they made one major error which eventually required a retraction and some other possibly erroneous assumptions.

First of all, Ming Pao decided to use the density of 100 persons per 100 square meters, in spite of the Lee's recommendation of 85 meters.  This resulted in the estimate of 72,000 persons per trip.

The major error was the same "tree planting problem" error made by the Civil Human Rights Front.  Ming Pao had made the non-physical assumption that the number of trips was 5.5 / 1.5 = 3.67.  It is now accepted that the bulk of the marchers left between 230pm and 545pm, so that there is only time for 2.17 trips.  At a density of 72,000, the number of marchers is 72,000 x 2.17 = 156,000.  I emphasize again this was Ming Pao's error and it was not the fault of Thomas Lee. 

In addition, Ming Pao assumed that the density of stationary people in Victoria Park would be the same as when they get on the road.  This is surely wrong.  I can stand right next to someone, but when I walk, I would need some more space to avoid collision or tripping unless we all walk very slowly.  Empirically, you can observe this phenomenon when you disembark from Star Ferry: the passengers are crowded together as the ferry approaches the pier, but they space out as soon as they start walking on land.  Alternately, you can compare a full parking lot against vehicular traffic in motion.  So the density appears to be overstated for the march.

Finally, the density of the stragglers at the tail end of the march is likely to be less than the front end.  By 6pm, the march route had been closed from 6 lanes down to 3 lanes in Wanchai and the same in Admiralty by 7pm.  Ming Pao had assumed a constant density, and this is another overstatement.


(Oriental Daily)  Satellite Photo Accurately Calculates The Number of Marchers  July 3, 2004.

[translation] The police and the Civil Human Rights Front had a large difference in estimating the number of marchers.  According to a satellite photo analysis expert who used a large-scale analysis to estimate the number of marchers on the basis of the road area, the total road area is about 72,000 square meters.  If each 100 square meters contain 85 to 100 persons, then the whole roadway would contain 72,000 if filled.  Therefore, he estimated that the number of marchers would be 220,000, which is about the same as the police estimate.

Lee Wei-peng is the general manager of a company that does land surveying, and he used an aerial photograph taken by a newspaper and applied a piece of software that is used for satellite photo analysis of large-scale geographical/environmental research and used the principles of diffraction to estimate the number of marchers.  He pointed out that the density of the crowd in the photo varied from sparse to dense.  Each 100 square meters contain 85 to 100 persons.  When a Victoria Park soccer field is filled, there are about 20,000 persons.

He said that the march route went from Victoria Park to Battery Lane in Central.  Subtracting the tram lanes that the police set aside for emergency vehicles, the entire road way is about 3,600 meters long and 20 meters wide.  The total area is 72,000 square meters.  If the entire roadway is filled with stationary people, without considering people who might have left or joined the march in the middle, the total number of persons on the roadway is 72,000.

He pointed out that the marchers must move forward, which would lead to spaces appearing in the road.  To accurate estimate the number of marchers, it would be necessary to take multiple photos from the start to the finish, including the starting and end points, in order to determine the march time and the number of waves of people.  If it is assumed that there were three waves during the period, then the number of marchers is estimated to be 220,000 which is about the same as the police estimate.

He pointed out that aerial photo analysis of march sizes is seldom done, but it has known accuracy.  He estimates that the cost for this estimation project was between $50,000 to $100,000.

He believes that the police error is unlikely to be subject to a lot of error.  But he thinks that the Civil Human Rights Front may have included those who left or joined the march in the middle, so it may not be surprising that they differ by twice as much.  He suggests that the two parties should describe their methods and compare.

衛星圖片準確計算遊行人數 新 聞

警 方 及 主 辦 七 一 遊 行 的 民 間 人 權 陣 線 , 對 遊 行 人 數 統 計 存 在 極 大 差 異 , 有 星 圖 片 分 析 專 家 以 專 門 為 大 型 規 劃 分 析 的 方 法 , 以 路 面 面 積 計 算 遊 行 人 數 , 他 指 整 條 遊 行 路 線 的 面 積 約 七 萬 二 千 平 方 米 , 每 平 方 米 約 八 十 五 至 一 百 人 計 算 , 填 滿 整 條 路 線 約 有 七 萬 二 千 人 , 估 計 遊 行 人 數 約 有 二 十 二 萬 人 , 與 警 方 估 計 相 若 。

本 身 是 土 地 測 量 師 的 星 眺 有 限 公 司 董 事 總 經 理 李 偉 鵬 , 日 前 利 用 報 章 一 幅 於 維 園 高 空 拍 攝 的 圖 片 , 以 專 門 作 大 型 規 劃 、 地 質 研 究 或 空 氣 污 染 的 星 圖 片 分 析 軟 件 , 用 多 光 譜 分 析 原 理 評 估 遊 行 人 數 。 他 指 由 於 圖 片 中 人 群 有 疏 有 密 , 每 平 方 米 約 有 八 十 五 至 一 百 人 計 算 , 填 滿 六 個 維 園 足 球 場 約 有 二 萬 人 。

他 表 示 , 整 條 遊 行 路 線 以 維 園 至 中 環 炮 台 里 , 扣 除 警 方 畫 為 緊 急 通 道 的 電 車 路 段 , 全 段 路 面 長 約 三 千 六 百 米 , 闊 約 二 十 米 , 整 體 面 積 約 七 萬 二 千 平 方 米 , 若 整 段 路 面 填 滿 不 動 的 人 群 , 未 計 中 途 插 隊 或 離 隊 人 士 , 最 多 有 七 萬 二 千 人 。

他 指 , 當 遊 行 人 士 向 前 移 動 時 , 便 會 有 位 置 出 現 疏 落 情 況 , 要 準 確 計 算 遊 行 人 數 , 必 須 從 高 空 以 垂 直 拍 攝 方 式 , 由 起 點 及 終 點 每 個 段 落 拍 攝 , 再 以 確 實 行 畢 的 遊 行 時 間 及 出 發 次 數 , 再 減 除 圖 片 中 的 空 隙 , 便 可 準 確 計 算 遊 行 人 數 。 以 全 個 遊 行 時 間 共 有 逾 三 程 計 算 , 推 算 遊 行 人 數 約 有 二 十 二 萬 人 , 與 警 方 估 計 相 若 。

他 指 以 星 圖 片 分 析 軟 件 統 計 遊 行 人 數 並 不 常 用 , 但 有 一 定 的 準 確 性 , 以 今 次 大 型 遊 行 計 算 估 計 只 需 花 費 五 至 十 萬 元 。

他 相 信 警 方 的 數 字 偏 差 度 不 大 , 但 認 為 民 陣 可 能 有 計 算 離 隊 及 插 隊 人 數 , 統 計 數 字 與 警 方 相 差 一 倍 亦 不 奇 , 建 議 雙 方 應 列 出 計 算 方 式 作 比 較 。


(Ming Pao)  Yip Siu-fei: 192,000 marchers by computer analysis.  July 16, 2004.

[translation]  The author and the team he led used two counting stations and a sample survey to estimate the number of 7/1 marchers.  We were looking for a more effective and accurate method of estimating the number of marchers.  The author's method consists of counting the flow within every minute out of 15 minutes, and then used that as the average of the 15 minutes from which we derived the total number of marchers.  At the second counting station, we also interviewed a sample of people in order to estimate the number of people who joined in the middle.  We added another 10% for those who joined the march by other means.  After adjustment, the total number of marchers was estimated to be 165,000 with a confidence interval between 140,000 and 190,000.

Ming Pao used a helicopter to take photographs which were analyzed by a satellite photo expert by computer software.  This is an important point of reference.  It was estimated that there were 72,000 persons on the road between Victoria Park and Admiralty.  According to Ming Pao, the total march lasted 5 hours and 30 minutes.  Since the trip took 90 minutes, there were 3.67 waves in total and the number of marchers was 264,000.

But like the Civil Human Rights Front, Ming Pao made the same error for the "tree planting" problem and double-counted the last trip.  If we began counting by the time that the first wave reached the Central Government Office, there was only 4 hours  If the trip took 90 minutes, then there were only 2.67 waves and not 3.67 waves.  The actually number of marchers is 192,000, and not the 264,000 that was reported.  Furthermore, based upon the author's observation, the 72,000 figure was obtained at the peak time (4:30pm).  After 6pm that day, the march route was reduced from 6 lanes to 3 lanes.  After adjusting for this, the number from the aerial photo analysis should be less than 192,000.

Since some media have commented about the discrepancy between the 164,000 from the author and his team and the 264,000 from the aerial photo analysis as compared to the 530,000 from the Civil Human Rights Front, the author believes that your newspaper should issue a correction.  All other newspapers and estimating organizations and individuals have published numbers that are less than 200,000, which is consistent with the police estimate.  We have reason to believe that the number of marchers is not more than 200,000 is supported by data.  The estimate from the Ming Pao aerial photo analysis is quite close to the estimate obtained by the author with human counting.

筆者及其領導的研究小組,以兩個點票站和一個抽樣問卷調查估計「七一」的遊行人數,希望能更有效和準確地估計遊行人數。筆者的點算方法以每15分鐘其中一分鐘的人流,為每15分鐘的平均人數,從而估計經過兩個點票站的人數,以及在第2個點票站進行抽樣調查,估計在這兩點間插隊的遊行人士數目。再加上10%的遊行人數,作為在其他地方插隊人士的估計。經過調整後,估計遊行總人數為16.5萬,可信度在1419萬之間。

《明報》以直升機拍攝、由衛星專家以電腦分析的數據,是一個很重要的參考材料。由維園至金鐘的路面上之遊行人數估計為7.2萬。根據《明報》的報道,以整個遊行歷時5小時30分鐘、每90分鐘為一轉作計算,全程共有3.67轉,而所得的遊行人數為26.4萬。

但《明報》的估計像民陣的一樣,犯了很容易錯的「植樹問題」,即將最後一轉遊行人士重複點算。所以若以第一轉遊行人士到達政府總部的時間起計,至整個遊行結束為止,只有4小時,每轉為90分鐘,則只有2.67轉,不是3.67轉,而得出之總遊行人數實為19.2萬,而不是所報道的26.4萬人。再者就筆者觀察,7.2萬之遊行人數是從當日最高峰的時段(4時半)所得,以及當日6時後,遊行隊伍由6行行車線縮減至3行。所以經調整後,電腦分析圖片所得出的遊行人數應比19.2萬還要少。

因為近期不同媒體的評論,以筆者及其研究小組所得之16.5萬和衛星圖片專家以電腦分析的26.4萬人數作為評論基礎,跟民陣的53.3萬作比較,所以筆者認為貴報有作出更正的必要。再看其他報章、點算遊行人數的機構和人士發表的遊行人數都在20萬以內,而警方的估計也與此數目一致,我們有理由相信「遊行人數不超過20萬」是一個有數據支持的點算結果,《明報》的衛星專家電腦分析與筆者用人手點算的結果也十分相近。


(Ming Pao)  Sorry, We Got The Number Of Marchers Wrong!  July 15, 2004.

[translation]  IT is two weeks since Hong Kong saw the July 1 march, but controversy persists over the number of people who joined it. It should fall between 200,000 (the police's estimate) and 530,000 (the figure provided by the Civil Human Rights Front, the organiser). It was reported yesterday in the press that Dr John Bacon—Shone, a Hong Kong University academic, puts it between 105,000 and 120,000.

The same day the present writer read Hong Kong University statistics and actuarial science senior lecturer Dr Paul Yip's contribution to this newspaper's Forum Page. It says the story titled "264,000 March According to Computer Analysis" which appeared on page A11 in the July 2 issue of Ming Pao contains a serious error. It says the number of the last "batch" of protesters was twice included. We must not only thank Dr Yip for pointing out this error but also set the record straight and apologise to our readers.

The long and short of the matter is as follows. We at Ming Pao consider the number of people taking part in a march to be quite newsworthy. However, over the years, the police's estimates have been at variance with and often vastly different from march organisers'. For example, the figure which the organiser of this year's July 1 march gave is more than 100% higher than the police's estimate.

To arrive at the truth we made new attempts. This year we rented a helicopter, from which our people took photographs of the procession right above it. We then asked Thomas Lee, a satellite photo analyst, to count marchers on the route between Victoria Park and the Central Government Offices (those in a "batch") using space technology — a piece of software for analysing satellite photos. He estimated that there were about 72,000 marchers in a "batch".

It took 90 minutes for protesters to march from Victoria Park to the Central Government Offices. The march lasted 5.5 hours. We reckoned there were 3.67 "batches". Therefore, 26,4000 people marched on July 1. However, there we made a "tree planting problem" error. We included the last "batch" twice. Our figure is 72,000 bigger than the true number, which should be 92,000.

Apart from Dr Yip, quite a few academics from Hong Kong University, the Chinese University, the University of Science and Technology and Baptist University have done much research to estimate the number of people who marched on July 1. By making such efforts, they have brought into play the civic spirit and the scientific spirit. They have provided the public with estimates that are relatively scientific and accurate. Those academics have employed different methods, which have their peculiar strengths and weaknesses. However, they have largely obtained relatively accurate estimates by scientific means. They have put the number within reasonably narrow neighbourhoods so that we have what is close to the truth. We ought to salute them.

The number of people who marched on July 1 this year matters. Of that the government's withdrawal of the Article 23 bill in the wake of the July 1 march last year is corroborating evidence. However, credibility matters more. It is wrong to overestimate or underestimate such numbers for any political reasons.

Fewer people marched on July 1 this year than last year. Still, that day about 100,000 citizens took to the streets, sweltering and soaked with sweat. They did express their dissatisfaction with the government and their aspirations for democracy loud and clear. Only by heeding the people's wishes, increasing democracy and improving the SAR's governance can the government allay public resentment.

March organisers' estimated numbers of participants do not seem objective, nor do the government's or the police's. That may unnecessarily arouse controversy. When organisers or the police make public their estimates, they should also disclose the methods they use and the sources of their data so that professionals, academics and citizens can form their own judgements.

We hope academics will persist in fulfilling intellectuals' functions and vigorously helping foster the civic spirit in Hong Kong. Their selfless commitment does underscore Hong Kong's strengths as a mature civil society and epitomise professionalism and the core values on which Hong Kong prides itself.

「七一遊行」至今兩星期,但遊行人數的爭議依然餘波盪漾、漣漪不絕。七一當天,遊行人數介乎警方估算的20萬人,至主辦單位「民陣」聲稱的53萬之間﹔據昨天的報道,香港大學學者白景崇公布的最新估算數字,是10.512萬人。

同一天,筆者看到香港大學統計精算系高級講師葉兆輝給《明報》論壇版的稿件(明天刊於論壇版),指出明報72A11版「電腦分析26.4萬人遊行」一文犯了一個嚴重錯誤──重複點算了最後一轉遊行人數。我們不單要多謝葉兆輝博士的指正,更必須鄭重更正,並向讀者致歉。

事情的來龍去脈是這樣的﹕《明報》鑑於遊行人數是新聞焦點,而多年來警方與主辦單位的數字常有出入,有時甚至相當懸殊,例如今年七一的數字就相差超過一倍。

為了尋找事實真相,我們作出了新嘗試,今年七一特別租用直升機垂直拍攝遊行隊伍,再委託衛星圖片分析專家李偉鵬,利用太空科技的衛星圖片分析軟件,分析出由維園至政府總部的全路段一程約有7.2萬人。

我們再按每90分鐘行畢一程,以遊行5.5小時推算出共有3.67程,得出遊行人數為26.4萬﹔然而,我們就是在這裏犯了「植樹問題」的推算錯誤,重複點算了最後一轉遊行人數,即多計了7.2萬人,實數應為19.2萬人。

除了葉兆輝之外,今次七一遊行的人數估評還有多位港大、中大、科大、浸大的學者積極參與研究,他們的努力,發揮了公民社會的力量和獨立的科學精神,使公眾對遊行人數有一個比較科學客觀的推算。雖然不同學者以不同方法進行統計,方法上各有利弊,但整體而言,已能比較科學地把估算人數縮窄至一個較合理範圍,使我們比較接近事實真相。謹向這些學者致敬。

遊行人數多寡是重要的,去年七一遊行後當局撤回第23條的立法,是一個佐證﹔然而,更重要的是誠信,如果為了政治原因,故意誇大或壓縮人數都是錯誤的。

今年七一遊行的人數雖比去年少,但仍有以十萬計的市民冒高溫,汗流浹背地上街,市民對政府施政的不滿,對民主政制的渴求,仍是清晰而響亮的。政府必須聆聽民意、發展民主、改善管治,才是消解民怨、對症下藥之本。

此外,鑑於主辦單位以至政府及警方對遊行人數作出估算,都會予人有欠客觀之感,引起社會上不必要的爭執。假如今後主辦單位或警方要公布遊行人數估算,便應全面公布計算方法及每一項數據的來源,以便專業學者及社會進行審計。

我們更希望學者能繼續發揮知識分子的作用,積極推動香港公民社會的發展。學者無私專業的奉獻,更突顯香港公民社會的成熟,並實實在在地彰顯了香港引以為榮的專業核心價值。

This official Ming Pao correction now assumes that the density of the crowd was constant between the departure times of 230pm and 630pm from Victoria Park (and corresponding arrival times of 400pm and 800pm).  It has now been accepted that the last significant departures had occurred by 545pm, so that the corrected figure is still an overstatement.  If you accept the 545pm time as the last departure, then the corrected figure is 2.17 x 72,000 = 156,000.

The above is a minor quibble.  Since Ming Pao disclosed what its assumptions and calculations were, the reader can make his/her own evaluation.  In any case, the behavior of Ming Pao has been honorable and exemplary to this point.