都市中的每個人無時無刻都發生著屬於自己的故事，而這些故事都會留下痕跡。有些人會將它以相片、影像、圖畫、文字、味道、甚至聲音等方式記錄下，但是大部分人的故事將會隨著時間推移而漸漸淡忘。都市中的「觀察者」不只僅限於我們人類，也有一些相對客觀的觀察者在默默紀錄著我們生活中的一切，例如監視器。它能連續且客觀地利用影片記錄我們每一天所發生的故事，並完整地保存下來。隨著影像辨識技術的發展與普及，讓我們的生活更便利，不但能取代人力，與監視器影像結合後，甚至能找出我們平時不易發現或無法察覺的都市涵構(context)。然而，這些記錄雖然完整、詳細，卻也需要大量的儲存空間。若我們能善用這些資訊，或將它們用另一種資料形式保存下來，就可對這些資訊做後續、長期的分析與應用。

我們提出了一套影像辨識轉換系統，目的是希望能夠使用一種新的都市觀察方式輔助做設計和決策，此系統將大量的影像(非結構性資料)轉為行人座標(結構性資料) ，不但能從影像資料中擷取我們所需要的資訊，減少大量的儲存空間，也能使資料更具有可讀性。我們進一步地將這些資料以視覺化的方式呈現，產生人流熱力圖。讓設計者能夠快速地透過熱力圖中的特徵圖形 (pattern) 了解不同場域的特性。透過這些特徵圖形，希望未來可做為機器學習的資料集，讓機器學習並歸類涵構資訊的圖形特徵，與大數據結合後進而形成涵構察覺系統，偵測環境的涵構資訊並給出合適的回饋。

台中文華路處於校園、住宅與商業空間的交界處，一天中的人潮分佈變化顯著，因此非常適合作為本論文的研究基地。夜市為文華路帶來不少商機，卻也帶來不少問題。本研究使用影像辨識技術結合監視器系統，偵測台中文華路群眾分佈與密度，根據周圍環境狀況分類出數種經常發生的事件與涵構特徵，並歸類出人潮分佈的特徵圖形(pattern)，進而針對文華路特徵圖形進行分析，提出文華路的空間治理建議，最後用設計回應這些空間治理建議。

Everyone in the city has their own stories happening all the time, and these stories will leave traces. Some people will record it in the form of photos, images, pictures, text, taste, and even sound, but most of the stories will be gradually forgotten over time. The "observers" in the city are not limited to us humans. There are also relatively objective observers who silently record everything in our lives, such as monitors. It can use the film continuously and objectively to record the stories that happen every day and keep them intact.

With the development of image recognition technology, our lives have become more convenient. It can not only replace manpower, but it can even identify urban contexts that we usually cannot find or cannot detect. However, although these records are complete and detailed, they also require a lot of storage space. If we can make good use of this information, or save them in another form of data, we can do further analysis and application of them.

Taichung Wenhua Road is located at the junction of campus, residential and commercial space. The distribution of people in a day changes significantly, so it is very suitable for this paper's research base. The night market brings many business opportunities to Taichung Wenhua Road, but it also brings many problems. This research uses image recognition technology combined with a monitor system to detect the distribution and density of the masses on Taichung Wenhua Road.
According to the surrounding environment, several frequently occurring events and context characteristics are classified, and the feature pattern of the crowd distribution is classified. We then analyzed and proposed the urban design method applied to the Taichung Wenhua Road.

[1] B. Schilit, N. Adams, and R. Want, “Context-aware computing applications,” Mobile Computing Systems and Applications, 1994. Proceedings., pp. 85 –90.

[2] M. Rodriguez, I. Laptev, J. Sivic, and J.-Y. Audibert. “Density-aware person detection and tracking in crowds,” IEEE International Conference on Computer Vision Workshops, 2011.

[3] N. Naik, J. Philipoom, R. Raskar, and C. Hidalgo. “Streetscore-predicting the perceived safety of one million streetscapes,” IEEE Conference on Computer Vision and Pattern Recognition Workshops, pages 779–785, 2014.

[4] P. Charalambous, I. Karamouzas, S. J. Guy, and Y. Chrysanthou, “A data-driven framework for visual crowd analysis,” Comput. Graph. Forum, vol. 33, no. 7, pp. 41–50, 2014.

[5] S. Ren, K. He, R. Girshick, and J. Sun, “Faster R-CNN: Towards real-time object detection with region proposal networks,” Proc. 28th Int. Conf. Neural Inf. Process. Syst., 2015, pp. 91–99.

[6] S. Yi, X. Wang, C. Lu, and J. Jia, “L0 regularized stationary time estimation for crowd group analysis,” IEEE Conference on Computer Vision and Pattern Recognition Workshops, Jun. 2014, pp. 2219–2226.

[7] Xu, Zhenning, Gary L. Frankwick and Edward Ramirez, “Effects of big data analytics and traditional marketing analytics on new product success: A knowledge fusion perspective,” Journal of Business Research, Volume 69, Issue 5, May 2016, pp. 1562–1566.

[8] 觀察、紀錄的連續性圖, https://studiotwo.files.wordpress.com/2007/04/sitephotos01.jpg

[9] 涵構察覺, https://en.wikipedia.org/wiki/Context_awareness

[10] Place Pulse網頁與資料收集, https://coolhunting.com/tech/place-pulse/

[11] Place Pulse 2.0, https://www.researchgate.net/figure/Image-examples-with-their-perceptual-attributes-from-all-six-continents-from-the-Place_fig4_305983690

[12] Streetscore學習街景中的圖像特徵, https://www.pnas.org/content/114/29/7571/tab-figures-data

[13] 影像辨識, https://zh.wikipedia.org/wiki/%E8%A6%96%E8%A6%BA%E5%BD%B1%E5%83%8F%E5%81%B5%E6%B8%AC%E6%8A%80%E8%A1%93

[14] Landing AI 社交距離檢測工具, https://landing.ai/landing-ai-creates-an-ai-tool-to-help-customers-monitor-social-distancing-in-the-workplace/

[15] 大數據分析與傳統行銷分析,
https://medium.com/marketingdatascience/%E8%B3%87%E8%A8%8A%E7%A7%91%E6%8A%80%E5%A4%A7%E9%9D%A9%E5%91%BD-%E5%A4%A7%E6%95%B8%E6%93%9A%E5%88%86%E6%9E%90%E8%88%87%E5%82%B3%E7%B5%B1%E8%A1%8C%E9%8A%B7%E5%88%86%E6%9E%90-8793ac4b5832

[16] 地圖也能參一“咖”, https://storymapcontest.maps.arcgis.com/apps/MapJournal/index.html?appid=9ecf416315f848f89dfb5b2eccbd29f9

[17] Google Map, https://www.google.com.tw/maps

[18] 「Code Brew Project」AI 咖啡, https://www.foodytw.com/article/87457

[19] Preface Coding, https://www.preface.ai/

[20] 自然語言處理(NLP), https://zh.wikipedia.org/wiki/%E8%87%AA%E7%84%B6%E8%AF%AD%E8%A8%80%E5%A4%84%E7%90%86

[21] Humans Process Visual Data Better, https://www.t-sciences.com/news/humans-process-visual-data-better

[22] Anaconda, https://www.anaconda.com/products/individual

[23] Spyder, https://www.spyder-ide.org/

[24] Python, https://www.python.org/

[25] Seaborn, https://seaborn.pydata.org/

[26] 類神經網路, https://zh.wikipedia.org/zh-tw/%E4%BA%BA%E5%B7%A5%E7%A5%9E%E7%BB%8F%E7%BD%91%E7%BB%9C

[27] 類神經網路架構, https://www.researchgate.net/figure/Artificial-neural-network-architecture-ANN-i-h-1-h-2-h-n-o_fig1_321259051

[28] GluonCV, https://cv.gluon.ai/

[29] Distributed (Deep) Machine Learning Community, https://github.com/dmlc

[30] Kepler.gl, https://kepler.gl/

[31] Kevin Lynch,《The Image of the City》(1960).

[32] 台灣土地使用分區顏色圖, https://maps.nlsc.gov.tw/pro/get_map_message.jsp

[33] 逢甲夜市隔音牆, https://www.peopo.org/news/48565