Person Localization and Crowd Sourcing in Aerial Images
Accurate detection and counting of persons are essential in natural disaster management and security applications including crowd management, detecting and counting people from flood or earth quick affected areas, convoy protection and tracking a ground-moving target (GMT) autonomously and so o
2025-06-28 16:28:46 - Adil Khan
Person Localization and Crowd Sourcing in Aerial Images
Project Area of Specialization Artificial IntelligenceProject SummaryAccurate detection and counting of persons are essential in natural disaster management and security applications including crowd management, detecting and counting people from flood or earth quick affected areas, convoy protection and tracking a ground-moving target (GMT) autonomously and so on.
Pakistan has exposed the most damaging natural disasters like earthquakes and reoccurring floods in the recent past. It is very difficult the detecting and counting of peoples in remote flooded or earthquake-affected areas. Search and rescue (SAR) teams face problems in search and provision of aid to the people who are in danger or forthcoming danger. Relief a recovery operation can be effectively conducted if the location and number of affected peoples are timely available. So, we decided to work on detecting person, localization, and recognition in UAV images. In this way, we can save many lives in natural disasters.
Pakistan contains a large border of 6,975 km (excluding the coastal areas). Pakistan is bordered by India to the east 3323 Km in length and bordered by Afghanistan to the northwest 2430 Km in length. Indian army continuously involved in a ceasefire violation from the eastern border. Afghan forces and nonstate actors (terrorists) attacks on Pakistan’s security forces. Terrorists also cross the border to attack military and civilian installations. Therefore, we need continuous border patrolling and surveillance. Loss of human life is being observed during border patrolling. So, surveilling and monitoring of border using UAV is effective. We can detect, localize and recognize the persons by using images/ videos taken by UAV from a station.
This project focuses on the person detection and counting in a clutter background. The main components of the project will be:
(1) UAV Platform for video and image acquisitioning,
(2) Data link for transferring of image from UAV to processing system,
(3) Development and use of algorithms for image enhancement, features extraction for human detection and recognition, and performance comparison of algorithms,
(4) Implementation on Raspberry-PI that can be used as onboard processing on UAV,
(5) Collection local environment large dataset for better accuracy.
The proposed system can be used for security surveillance, crowd management, and delivery services.
Project ObjectivesThe main objectives of our project are:
- Using different detection and recognition algorithms for the detection of a person
- Count the several people present in a specific area
- Performance comparison of different detection and recognition algorithms and obtain the best detection and recognition algorithm for detection of person
The proposed architecture is described in figure.
The proposed project block diagram as shown in figure above is divided into two parts: one is user-controlled and the other is the implementation of a person detection and recognition algorithm.
First, take videos with the help of UAV (drone). Users fly UAV (drone) manually with the help of remote control and take video from a specific area where crowd present.
In the second part perform different detection and recognition algorithms in raspberry-pi for the detection and counting of several persons.
- First, separate frames (images) from video and apply different pre-processing and image-enhancement methods for the removal of noise. For the removal of noise, used resizing, filtering and edge sharping.
- With the help of cropping, we extract an area of interest. In our project area of interest in those areas in which a person’s present.
- Extracting features of an object in our case extracting features of a person’s.
- Train model with the help of different training algorithms. For the training model making 20k images dataset and then train this dataset with the help of training models. Testing a trained model with the help of a test image and then count several persons present in a specific area.
- After the detection and counting number of person display output on the monitor screen (detected and counted several persons).
Accurate detection and counting of persons are essential in natural disaster management and security applications including crowd management, detecting and counting peoples from flood or earth quick affected areas, convoy protection and tracking a ground-moving target (GMT) autonomously and so on.
Pakistan has exposed the most damaging natural disasters like earthquakes and reoccurring floods in the recent past. It is very difficult the detecting and counting of peoples in remote flooded or earthquake-affected areas. Search and rescue (SAR) teams face problems in search and provision of aid to the people who are in danger or forthcoming danger. Relief a recovery operation can be effectively conducted if the location and number of affected peoples are timely available. So, we decided to work on detecting person, localization, and recognition in UAV images. In this way, we can save many lives in natural disasters.
Counter-terrorism operation is being conducted by law enforcement agencies in remote areas. Detection of terrorists in a remote area is very necessary for the safe journey of the convoy. Tracking of a ground-moving target (GMT) autonomously is helpful to law and enforcement agencies. Surveillance of defense installations and security setups is very necessary for the current situation. Surveillance, localization, and recognition of persons will help encounter enemies far away from security systems.
Disaster management:
Crowd management
Detecting and counting peoples from flood
Earth quake affected areas
Security applications:
Convoy protection
Tracking a ground-moving target (GMT)
Surveillance
Technical Details of Final Deliverable-UAV Video/Image Acquisitioning Platform
Camera Resolution 1280x720
Camera Resolusion 640x260
ON-board and ground control system for UAV
-Software for image enhancement, features extraction, and human detection in images
For Image enhancement using Gussian, sobal and canny edge detector.
For features extraction and human detection using Haar- Cascade, Tamplet matching, YOLO and CNN.
-Raspberry-PI based processing platform that can be ultilized both at onboard and ground system
Lunix based operating system
Used for Image Processing, and real time person detection and regonition.
Final Deliverable of the Project HW/SW integrated systemCore Industry SecurityOther Industries Transportation Core Technology Artificial Intelligence(AI)Other Technologies Internet of Things (IoT)Sustainable Development Goals Sustainable Cities and Communities, Life on LandRequired Resources| Item Name | Type | No. of Units | Per Unit Cost (in Rs) | Total (in Rs) |
|---|---|---|---|---|
| Total in (Rs) | 80000 | |||
| 450 quadcopter frame | Equipment | 1 | 2800 | 2800 |
| Quadcopter legs | Equipment | 4 | 150 | 600 |
| Radiolink T8FB | Equipment | 1 | 8500 | 8500 |
| APM 2.8 flight controller | Equipment | 1 | 16000 | 16000 |
| Gens 4000 45c 3s LiPo battery | Equipment | 1 | 9500 | 9500 |
| 30A Escs | Equipment | 4 | 1400 | 5600 |
| A2212 1000kv motors | Equipment | 4 | 1300 | 5200 |
| B6 Charger | Equipment | 1 | 3100 | 3100 |
| 10 X 4.5 props set | Equipment | 2 | 400 | 800 |
| Rf7 simulator | Equipment | 1 | 2600 | 2600 |
| Power module for Raspberry Pi | Equipment | 1 | 1800 | 1800 |
| Raspberry Pi | Equipment | 1 | 9500 | 9500 |
| Pi camera | Equipment | 1 | 4000 | 4000 |
| Wires, Soldering Iron, Shipping, Transport, Thesis Printing | Miscellaneous | 5 | 2000 | 10000 |