Introduction - person-in-hangang/HanRiver GitHub Wiki

Motivation

Currently, Han River is also equipped with alarms to detect people who jump into the river. However, the alarm often detects objects that are not human and causes malfunction. And the alarm only recognizes people hanging around the bridge for a long time and alerts. Therefore, the controller should judge whether a person actually falls. We also don't know when people have fallen. Since the alarm does not even know the situation under the bridge, it is also unknown where a person fell.

So we implemented this project to solve these problems.

This project

1. increased reliability by allowing detector to detect only people.
2. It also automatically detects and notifies people of falling even if they don't monitor them 24 hours a day.
3. In addition, it also provides information on the time when a person fell
4. and predicts where a person fell through tracking.

Introduction

The fall detection & drown prediction project was designed to help detect people falling from the Han River bridge quickly and rescue people from the water. So, when making and experimenting with the project, the location of the experiment was assumed to be the Han River Bridge. In the project, two Android devices were used to serve as cameras and an Android device to launch the application. The cameras supposed that each installed above and below the Han River bridge. And the application will be provided to 119 rescue teams. Cameras and applications are connected to one server.

The camera on the bridge (Camera 1) captures the situation on the bridge. Camera 1 has a blue line that marks the bridge railing. In this project, the line is drawn from the top right to the bottom left. The left side of the line is assumed to be a bridge and the right side is assumed to be the Han River. This line distinguishes whether a person has fallen or not. If a person crosses the baseline and is detected on the right after a person is detected on the bridge, it is recognized that a person fell off the bridge at this time. When a person falls, Camera 1 sends a picture to the server immediately. The server measures the height and find characteristics by extracting people from the photo received from Camera 1. After measuring the height, server determines whether the person who fell is an adult or a child. Also, it chooses the 3 most prominent characteristics among the various characteristics. The server then sends the application the time of the fallen and the predicted fallen person information.

While a person is falling, the camera under the bridge (Camera 2) tracks the person falling. Camera 2 has a detector that detects people, so it automatically recognizes people. Camera 2 is tracking people who fall using a bounding box, and if the bounding box is small like the predetermined size, Camera 2 thinks the person fell into the water. At this time, Camera 2 does not only recognize drowning, but also identifies the path of a person falling. Collecting bounding boxes from when a person is recognized until person falls into the water, it shows the path of falling. The bounding box when a person is first recognized is marked in red and the bounding box when drowning is marked in blue. When tracking is completed, Camera 2 sends a picture showing the trajectory to the server. The server immediately sends this picture to the application.

The navigation bar at the bottom of the application screen has three menus. The screen corresponding to the notification button is the main screen, and information about people falling from the aforementioned bridge is exposed. The screen is divided into three blocks, and in the first block, a picture of a person immediately after person falls is displayed. The user can then see the photo and see who has fallen. In the second block, the time of the fallen and the predicted fallen person information are displayed. In the third block, a photo showing the path of a person's fall is displayed. The user can predict where the person fell by looking at the picture. At the bottom of the block, you can see the latitude and longitude of where the camera is installed. After checking all the information, press the reset button at the top of the first block and all the information displayed on the screen disappears. So user can get new information again. At the top of the third block is a button called Map View. When user presses this button, user can check the location of the camera on the map. By pressing the graph button, you can see a line graph showing the status of fall over the past six months and a pie graph showing the status of fall by bridge. These graphs were created by taking information from a database that stored the fallen date, fallen time, and fallen location of the bridge where the camera whenever the projection was detected was installed. User can set up notifications on the Settings button screen. Even if the application is not turned on, push alarms can be set so that people can know when they fall.