Bangladesh is frequently hit by many natural disasters, particularly cyclones, floods, mud slides, and drought. The country's approximately 710 kilometers of coastline leave huge tracts of land open to the destructive effects of cyclones and storm surges. The cyclones generate surges up to a height of several meters which sweep over the flat coastal region. It damages lives, properties including crops and infrastructure etc. Since late 1960s, remote sensing is being used for monitoring cyclones in Bangladesh. Early Warning System has been established in the country, which effectively reduces the damage to life and property. The researchers in Bangladesh believe that effective search and rescue operation can further reduce the death toll. During severe cyclone, one of the biggest challenges is that it takes several days for the aid workers to gain access to some of the hardest-hit areas. The roads might be blocked due to the fallen or broken trees, which take time to clear out. Some roads are also obstructed by flood waters, which have to recede. Under these circumstances the search and rescue (S&R) operation is greatly hindered and timely aid does not reach to the victims. As a consequence, death toll might increase. Ministry of Food and Disaster Management (MoFDM), in association with the Bangladesh Red Crescent Society (BDRCS), is implementing Cyclone Preparedness Programmes (CPP) in the 12 coastal districts of the country to minimize loss of lives and properties in cyclone disaster. Bangladesh Air Force (BAF) also plays a vital role in the disaster management in all stages. During emergency response stage, only BAF conducts aerial surveillance for S&R operations. The usual approach for this purpose is to use manned aircraft equipped for covering wide area with special sensors and to assign the actual recognition task (surveillance) to the crew. However, in the usage of manned aircraft, it is difficult to operate from low altitude. A binocular telescope is usually employed in the manned aircraft for the magnification to detect small targets from high altitude. In that case, the range of vision for searching becomes narrow and the possibility of oversight must increase. On the other hand, S&R operation from ground is conducted by NGOs, Bangladesh Military, and other government organizations with the aid of community involvement. Different organization with their own accessories, works as different group in order to carry out their own disaster response activities. But these operations are often hindered and delayed significantly due to inaccessibility to the area caused by damaged infrastructure or due to the lack of resources. Also it is very difficult for the ground workers to predict the location of the victims unless they have detailed spatial information of the disaster scene. As a consequence, the S&R operations may be conducted where it is unnecessary or maybe omitted where it is must. Hence the possibility of not being able to identify victims might increases. During severe cyclone it is impossible to conduct aerial surveillance for the entire coastal area of Bangladesh with the limited resources of BAF. The other approach might be to provide the ground teams with detail spatial information about the disaster scene so that the search area can be narrowed down and if possible pinpointed. In that case, the rescue commander would be able to allocate resources with more efficiency. And the success rate of search and rescue operations might increase.

Aerial images can play a vital role in this case. The usual methods for acquiring aerial images are conventional aircraft or the satellite systems. But both of these systems have their own limitations. Acquiring aerial imagery using conventional aircraft during a disaster can be difficult when there is a lack of suitable runways. Such situation might occur if the runway is damaged by the disaster or the disaster area is far away from a suitable runway. Satellite systems can also provide imagery and have proven valuable in widespread disasters such as hurricane, flood, forest fire etc. Unfortunately, the temporal and spatial resolutions of satellite systems make them relatively ineffective for emergency response. Unmanned Aerial Vehicle (UAV), as shown in the figure below, can be effectively used for capturing aerial imagery. A properly equipped UAV can cover a large area with varied types of sensors in order to capture detail spatial information of the disaster scene. Because of the small size of the system it is possible to fly close to the area of interest and capture high resolution images using low cost digital cameras. Since UAV is not heavy, it can be carried to the disaster scene and can be operated as on need basis.

The objective of this research is to develop a UAV based tool to enhance the existing S&R operation in Bangladesh. This tool, which is intended to be used by the ground S&R teams, would be able to captures high resolution images of the disaster area as on need basis and to cover a large area within a short period of time. The tool should also be able to handle and process the acquired data in order to pinpoint the search area as well as to detect victims from image sequences processing. The tool would provide support for visual inspection of the raw and processed data, which would help the disaster commander for effective disaster response and efficient resource allocation. Successful S&R operation based on this tool lies in the fact that, it must maintain a specific time frame for data capturing, processing, and interpreting. The proposed Search and Detection (S&D) tool can be incorporated within the existing S&R operation framework in order to make the existing disaster emergency response by the ground workers more effective. The tool must possess the following specific characteristics:

Convenient →easy to use and operable in the disaster area

Fast →in acquiring spatial information

Reliable → in pinpointing search area and locating victims from the acquired spatial information.

Cheap →so that it is affordable for individual union (group of villages) in the remote marine islands.

From the application point of view, the S&D tool is divided into the following three components:

a)Data Acquisition System: This system consists of the

・UAV platform

・Sensor system:

i.For capturing spatial information.

ii.Includes GPS and Camera (still and/or video) etc depending on the application type and the payload constraint of the UAV.

・Synchronized data acquisition system and data storage

b)Data Processing System: The data processing system, named as "DataViewer" provides a faster, logical and accurate means of handling and processing a large amount of spatial data acquired by the UAV system. This component deals with the processing of the data in order identify the objects (victims) from image sequences and compute its real-world location.

c)Image Browser: This component provides a graphical user interface for browsing the raw as well as the processed images. It helps the rescue commander for visual inspection of the processed spatial information in order to narrow down the search area for the ground workers.

In order to implement the proposed UAV based tool within the existing disaster management framework, we have focused on the 'Disaster Preparedness' and 'Emergency Response' phases of the disaster management cycle as shown in the following figure:

During preparedness phase, as part of public awareness program, the people of the target area would be requested to wear a specific color (bright orange for example) dress during early warning period of cyclone. This specific color can be easily identified in the images by visual inspection as well as by automatic image processing. In this phase, UAV flight plan for data acquisition is also established based on the respective area maps which can be explained with the following diagram:

The flight design parameters include flight height of the UAV: in order to achieve a certain ground resolution in the acquired images, camera exposure interval: in order to maintain a specific overlap among the consecutive images in a strip, and location for take-off and landing of UAV considering intermediate refueling in order to cover a specific size of area. After the cyclone hit, UAV is deployed for data acquisition of the disaster area. The acquired data is then processed with DataViewer for detecting objects based on specific color information as well as for computing their real-world location. Visual inspection of the processed data helps the rescue commander in narrowing down the search area as well as in allocating the resources. The conceptual framework of the S&R operation implementing the S&D tool is shown in the following figure:

The blue bounding box in the above figure represents the extent of our research focus i.e. the S&D tool within the proposed UAV based S&R operation.

Several experiments have been conducted with different UAV platform and different sensor system. The acquired data are efficiently handled and processed with DataViewer for data quality checking, data preparation for computing absolute orientation parameter of images, object detection by image sequence processing, computing the real-world location of the detected objects, and finally for visual inspection of the acquired. The experimental results clearly demonstrate the high applicability of UAV borne sensor system for capturing very high resolution images. Data processing results with DataViewer shows that the acquired data can be effectively used to generate very detail and accurate spatial information, which in turn can be used for effective disaster monitoring i.e. effective emergency response. It is also evident that we can easily overcome the limitations of Satellite Images in case we would want to use those for emergency response purpose.

Due to the UAV flight safety regulations in Japan, the test flights for our experiments are restricted to unpopulated areas such as the riverside or mountainous area. Usually such places are very small in size as compared to the real disaster area. As a result the actual flight design parameters except the 'UAV Flight Height' could not be implemented. Flight height is a crucial parameter in order to represent a specific object in images with a certain number of pixel counts. Such requirements are sometimes very important in order to detect objects by automatic image processing. For our future work, we need to conduct a real scale experiment in order to develop a scenario for acquiring and handling real volume of data as well as to validate the full capacity of the DataViewer. Such experiment would also allow us to implement actual flight design parameters such as UAV flight speed, image acquisition interval, and area coverage.

Figure: RPH2 UAV, a Product of Fuji Heavy Industry

Figure: Conceptual Framework for Disaster Management

Figure: Establishing Flight Planning and Campaigning for Specific Color Dress

Figure: Implementation of S&D Tool within the Proposed S&R Framework

バングラデシュはガンジス川、ブラマプトラ川の河口に拡がる広大な低平地に位置し人口密度も非常に高いことから、サイクロンなど大きな災害に襲われるたびに大きな人的被害を生じている。1970年には30万人の死者を出し、2007年にも約3400人が亡くなっている。早期警戒情報の提供や避難シェルターの整備などにより死者の数は大きく減少しているものの、被害はいまだ甚大であり、さまざまな対応策が模索されている。その一つに災害直後の被災者の探索・救助活動がある。バングラデシュの沿岸部には多くの広大な中州があり、漁民などが多数居住しているが、陸上の交通路はほとんど整備されておらず、災害後の被害状況の迅速な把握、被災者の救助・救難活動は非常に困難である。先進国では一般に軍や救難担当組織がヘリコプターなどを動員し、捜索を迅速に展開するが、バングラデシュでは捜索すべき地域の広大さに比べて、軍の保有するヘリコプターの数は非常に限られており、迅速な捜索・発見は容易ではない。そのためNGOなどがその捜索・救助活動を補完的に実施している。具体的には数名程度のチームに分かれて数平方kmから10平方kmくらいの地域を担当し、ボート、徒歩などにより行っている。そのため探索効率は非常に低く多大の時間を要している。

近年、無人飛行体(無人飛行機や無人ヘリなど。UAV:Unmanned Air Vehicle)の低廉化、性能向上に伴い、カメラなどを搭載し地上の詳細な画像を大量・迅速に収集する試みが行われつつある。特に地震や火山の噴火といった大規模災害や斜面崩壊など、人が近づくことが困難な災害において、被害状況の早期把握のために試験的にではあるが利用され始めている。しかしながらバングラデシュのような開発途上国で比較的広い領域を対象に、現地での捜索・救助活動やその実施プログラムと密に連携した形で、被災者の探索・発見を目的としてシステムを設計・開発している例はない。

本研究はバングラデシュでの利用を念頭にしながら、少人数のチームが広い領域をできるだけ効率的に捜索し被災者を発見するシステムを、無人飛行体をベースに設計し、それを実現する技術開発を行うことを目的としている。

本論文は全体で7章からなっている。第1章は序論であり、バングラデシュにおける災害の歴史、災害対応における迅速な対応の必要性が整理され、研究の目的やそのオリジナリティが述べられている。

第2章は災害対応の現状と将来動向が整理されている。すなわちバングラデシュにおける災害対応の現状、衛星を中心としたリモートセンシングの利用の可能性と限界、サイクロン対応プログラム(CPP)と軍の役割、また遠隔地域におけるNGOの活動プログラムが整理され、NGOの探索・救助活動の中で何が必要なのかが具体的にまとめられている。

第3章は無人飛行体(UAV)をベースとした探索・発見システムの提案であり、既存のUAVシステムとの比較を行いながら、目標となる探索地域の広さ、NGO活動との連携のさせ方などを前提条件として、必要費用、データ処理の自動化レベル、処理時間制約、UAVシステムのハンドリングの容易さなどについて開発要件を明らかにしている。その上でシステム全体のデザインを行っている。またサイクロン対応プログラムの一環として、地域住民にオレンジ色の救難ジャケットを配布するなどの提案もなされている。

第4章は、データ取得システムの全体像であり、探索・発見システムの全体構成のうち、特にデータ取得部分に限ってより詳しく内容を述べている。すなわち搭載するセンサ(カメラ、GPS)やデータの形式、その特性などである。

第5章は、データ処理の中核になる「データビューワ」の開発であり、その構成とデータの処理フロー・アルゴリズムを述べている。「データビューワ」は、飛行デザインモジュール(対象地域の広さ、形状、取得画像の地上分解能や画像間のオーバラップ量などから飛行高度、速度、ルートなどを決めるモジュール)、データ処理モジュール(GPSデータから飛行軌跡を概略推定し、隣接画像データの自動的な接続を実現したうえで、さらに軌跡や傾きを精密に推定する機能を実現)、画像処理モジュール(推定された飛行軌跡データから3次元マップやオルソ画像作成を行う機能を実現)、画像判読モジュール(オレンジ色の救難ジャケットの自動発見などを目視判読と組み合わせ、被災者の発見を支援する機能を実現)などからなっている。

第6章は、データビューワの応用であり、UAVシステム上に実装された探索・発見システムの機能検証実験を行った結果が述べられている。国内ではUAVの飛行可能地域が限られていることから、河川やダムサイトなどを利用して実験が行われた。その結果、想定された要件をほぼ満足できる見通しが得られた。

第7章は考察と結論、今後の課題であり、現地において実際の探索・発見活動と連携した本格的な実証実験を行い今回開発されたプロトタイプを一層改良すべきこと、精密軌跡推定の処理時間の一層の短縮が必要なことなどが課題として挙げられている。

以上まとめると、開発途上国(特にバングラデシュ)における広域・激甚災害を対象に現地での対応活動プログラムなどと連携して機能するUAVシステムのプロトタイプを開発し、併せて救難ジャケットの配布など災害対応プログラムそのものにも踏み込んだ現実的な提案を行っている。これはリモートセンシング工学、測量学の新たな利用分野を開拓したものとして高く評価される。よって本論文は博士(工学)の学位請求論文として合格と認められる。