I have written several articles on the the topic of computer vision and importance of of implementing it. In this article, I am discussing about the importance of of implementing computer vision on modern UAVs for more robust autonomous operations.
Artificial Intelligence:
Artificial Intelligence which commonly known as AI, is simply the area of research that enable machines/computers to mimic natural human intelligence. The main purpose of AI is to give machine the ability to make self-decisions by analyzing the data by themselves, just like humans making decisions. There are many subfields of AI that are several subfields of Artificial Intelligence.
Computer Vision:
Computer vision is one of the major subfields of AI. The computer vision is simply the process of giving machines the ability to understand the surroundings and physical world by analyzing visual input data.
look at the following figure.
The previous figure is a primary example for using computer vision-based applications for analyzing visual input data and making decisions based on the data patterns. The first photos is a normal image that was taken on a street. As humans, we can analyze the photo with our eyes and make conclusions that there are people on the street, there is a color light and there are cars. But for a machine to understand it there should be a similar process to analyze the data and make decisions. This process is called the computer vision.
The computer vision can be implemented in modern UAVs and used for various applications. These applications include:
- Fire and smoke detection
- traffic monitoring/control
- security and surveillance
- wildlife/livestock monitoring
- agricultural activities
- media and entertainment
SKYCEY Technologies:
SKYCEY technologies is a AI-based tech start-up company that is focusing on building AI-integrated and fully autonomous unmanned vehicle systems to perform several complex tasks in multiple sectors with higher accuracy, efficiency, safety and productivity.
We are focusing on building:
- UAVs – (Unmanned aerial vehicles)
- UGVs – (Unmanned aerial vehicles)
- UUWVs – (unmanned underwater vehicles)
Our products will be fully autonomous and AI-integrated. So, there wont be any need for human or manual operating.
We are not just another drone company. We are building fully autonomous and AI-integrated unmanned vehicles for general purposes.
Areas of application for our products/UAVs:
Here are some of the AI-integrated detection systems that have been completed so far. These software programs can be implemented into the onboard computing unit of our UVs and onboard flight computer will perform the desired task by self-decision making and data analyzing.
Use Case – 1: wildlife / live stock monitoring:
wild life is a hard activity since there is a large area to cover. this would cost so much time and other resources such as cost to complete the monitoring manually by humans.
With our modern autonomous drones and their onboard computer units, we are implementing new technologies for wild-life and live stock monitoring.
The following figure shows how we are using artificial intelligence and computer vision to modernize the animal monitoring and livestock monitoring. ]
We are using our custom training models for detecting animals in real time and monitoring them for a specific period of time. Our models have an accuracy of 90-95%. The average accuracy of the YOLOv8 pre-trained model is 0.85. We are using the latest yolov8 custom-trained models as data models for detection and the onboard flight computer unit will analyze data in real-time and make self-decisions by itself. This process is called the edge-computing.
One of the major advantages of this is we don’t need a ground station to analyze data and make decisions, based on them which requires a lot of time. Our drones will complete the data collecting, analysis and evaluation while airborne in less than one minute. Furthermore, we are designing our drones to track and follow animals based on user commands.
This drone applications can be used for several purposes including:
- research and scientific purposes
- wild-life preserving activites
- wild-life protection and security
- live-stock monitoring for farmers
- protecting endangered animal species
- prevent illegal hunting
- prevent animal attacks such as bears, tigers by knowing their areas in advance
- warning hikers and other people who enter reserves or forests about suspicious animal activities and areas in advanced
- detect any injured, ill animals and respond quickly
Use Case – 2: Fire and Smoke detection:
Fires are very dangerous and harmful to human lives. They can cause huge damage to infrastructures as well. The most recent example is the devasted wildfire disaster in Los Angeles California. According to reports, The fires have caused at least 29 deaths, a similar number of injuries, the destruction of more than 16,000 structures, the damage or destruction of over 50,000 acres and thousands of evacuations. These wildfires are likely the worst in the history of California and the United States.
Our autonomous drones are built to detect any unusual fire in real-time even as a little smoke. our advanced deep learning models will analyze live data within milli-seconds and detect any small fire or smoke in real time. our drones can customized as fire-responding autonomous UVs. They will detect and inform human operators where the fire is, status of fire, and live visual outputs to make the responding more easier and faster. Our are designed to spray water as first line of responding to prevent small fires and preventing large fire damages.
Previous figure shows the difference between pre-trained normal model (left) and our custom-trained deep learning models (right). Our models are more accurate in analyzing and detecting fire and smokes from aerial view (bird’s eye) in real time. We are building more and more accurate detection models for real time monitoring.
Here are our test results for Live-fire detection from aerial view:
The following figure shows the clear difference of detecting by pre-trained and custom trained models when it comes to the priorities of detection.
On the left, the pre-trained detection detects humans and other common objects over fire. Also, the detection was only happened in more front layers of the image. The model is only looking for common data patterns that it can find first and show the result with detection. But with our custom-trained detection model and algorithm is analyzing every single layer of the live data feed and only looking for data patterns that identify fire or smoke. The drones can be customized for rescue operations that can detect fire and humans inside the buildings that caught on fire.
This drone application can be deployed for:
- wild-fire monitoring
- 24/7 monitoring in forests areas that identified as fire risk areas.
- urban patrols for any fire incidents
- essential equipment of local fire-departments
- for law-enforcement using
- first responders equipment
- large premises monitoring
- stopping wild fires
Use Case 3 – detect flying objects and collision avoidance:
One of the most dangerous scenarios for modern drones is colliding with other dynamic objects such as birds and other drones. specially for an autonomous UAV, self collision avoidance is a versatile feature. Computer vision is the major field of AI, that is used for obstacle avoidance. specially the dynamic object make a great threat.
With our advanced camera units and object detection algorithms, our drones will be able to detect these dynamic objects and change the path to avoid any potential collide within less than 1 second.
this figure shows that our custom trained models and deep learning algorithms can identify dynamic objects such as birds within less than 3 milli-seconds to respond. The onboard computer will analyze the data and decide whether to change the path or not.
Following figures show the test results for our algorithms for detecting dynamic objects for potential obstacle avoidance.
Use Case 4 – Path detection:
Computer vision is a major technology when it comes to path planning for autonomous vehicles. FSD cars including Tesla are using advance cameras to detect obstacles and path planning.
We are using advanced machine learning algorithms to detect static obstacles and plan paths by avoiding them with real-time data analysis. Here is how we do it.
The previous figure shows how the obstacle avoidance system is detecting the static obstacles and mark them in green color. then the system is checking unmarked areas in the frame to decide where should the vehicle go within next few seconds.
This path planning is essential for indoor autonomous operations since it can easily decide how to update the path by avoiding static obstacles.
Use Case 5 – Transportation and traffic control:
Our autonomous drone will be used for controlling traffic and analyzing the data that will be helpful for decision making. Here are a sample algorithms we are using for traffic control and analysis.
Our intelligent drone systems can identify incoming and ongoing vehicle in real time even though they are going on various speeds. With dynamic vehicle detection in real time, it is easy to detect, identify and keep record on vehicles that are important analyze the traffic on different places in various times with higher accuracy.
Previous figure shows how our advanced detection and classifying algorithms will analyze both lanes of a road in two specific areas that are marked in blue an d green rectangles. Then the algorithm will decide how many vehicles will cross these rectangle areas this analyze can help to understand the traffic on a road at any given time due to it’s ability to process within less than 5 seconds.
Our autonomous drones are able to monitor vehicle speed in real-time with higher accuracy. This will help law enforcement services to patrol the roads with minimum cost spending.
Our advanced algorithms and autonomous intelligent UAVs can be helpful for achieve several purposes in transportation including:
- real time traffic monitroing
- accident detection and responding
- traffic violation detection for law enforcements
- traffic prediction and route optimization
- Emergency or disaster management
- infrastructure inseption
- public transport systems monitoring
- environmental affecting and pollutions analysis
Use case 5 – Tracking and following specific persons throughout specific period of time.
This feature is specifically built for media production purposes. Yet this same detection system can be used in security patrols as well. In this feature, the drone will be locked into a specific person based facial recognition and then it will track and follow the person throughout a given specific time.
our advanced algorithms are capable of controlling drone movements based on the persons movement.
This features can be used for for media personal tracking in real time without any human operator by drone itself.
This feature can be also used for for security purposes such as tracking and following a suspicious person over the time.
Use Case 6 – Agricultural activities:
Agriculture is a major field that need modern technology to achieve more productivity and efficiency.
Or autonomous UAVs are being designed to detect crop for diseases, crop pest detection and control, crop status predictions etc.
With our autonomous drones, farmers can achieve higher productivity in pest control, detecting crop disease in early stages to reduce or prevent damages, crop monitoring daily basis, animal prevention from farm lands etc.
Use Case 7 – Object detection and tracking:
One of the major advancements of our intelligent drone systems are detecting specific objects from ground camera units and track them over the time. Our custom trained deep learning models can perform with more accurate detection results.
These features will help modern farmers to generate more productivity and efficiency and more profits with minimum cost spending.
Previously mentioned are the primary use cases for our autonomous and intelligent unmanned aerial vehicle systems.
SKYCEY is continuously building latest technologies by using Artificial Intelligence to improve autonomous technology and provide more efficient service to our clients and users.