DeepSky: AI as an Extra Pair of Eyes for General Aviation
1. Introduction
The growing air traffic, especially in areas with limited oversight, poses increasing challenges to airspace safety. Unlike commercial aviation, where every movement is strictly monitored, private aviation operates with greater freedom - introducing additional risks during flight maneuvers. While this autonomy is one of its greatest advantages, it also brings dangers, as pilots often rely solely on their own judgment and available tools.
A key issue is the lack of regulations enabling effective air traffic monitoring, particularly in smaller airports and aeroclubs. In major metropolitan areas, advanced surveillance systems track every flight, but such oversight is often absent in less populated regions. This significantly raises the risk of collisions, as private pilots do not always file flight plans, leaving their aircraft "invisible" to other airspace users. Additional hazards include the rising number of drones and even birds, which can unexpectedly cross flight paths.
In uncontrolled airspace, pilots are not required to report their routes. While they may do so voluntarily, submitting documentation at least two hours before takeoff is often seen as a time-consuming step. As a result, many choose not to comply, leading to numerous aircraft operating without any supervision.
To address this issue, regulatory measures should be implemented to require monitoring of all air operations - including private aviation. Without such reforms, collision risks will escalate alongside growing air traffic density.
Another major challenge is the limited availability of advanced collision-avoidance technologies, which are standard in commercial aviation. Small aircraft owners - typically operating planes valued at around 80,000-100,000 PLN - often cannot afford to install systems such as TCAS (Traffic Collision Avoidance System), which are commonly used in larger aircraft. The barrier is not only the cost - often equivalent to the aircraft’s value - but also the size and weight of such equipment. These devices take up considerable space and are relatively heavy, which poses an additional challenge for lightweight aircraft, negatively affecting aerodynamics and overall performance.
Compounding this issue is the difficulty in effectively tracking aircraft in the air. Flight-tracking applications available to pilots and aviation enthusiasts do not always display all aircraft currently in flight. The reason is that not all aircraft are equipped with modern tracking systems such as ADS-B (Automatic Dependent Surveillance-Broadcast), which enables real-time position transmission. Older aircraft models, in particular, often lack this technology, and while retrofitting is possible, it is a costly investment that not every pilot or private aircraft owner can afford.
2. "Extra Eyes" for Pilots
In response to all these challenges, the duo from BlockyDevs - Maciej Malik and Bartosz Solka - along with Michał Pstrąg, developed the innovative DeepSky application, significantly enhancing safety in general aviation. The idea for DeepSky was born from their own experiences and understanding of the everyday problems pilots face.
During February’s inaugural ElevenLabs Worldwide Hackathon, the Polish team outperformed ≈ 130 competitors to win 1st place in Warsaw. Using ElevenLabs' technology, the Lovable framework, and OpenAI’s ChatGPT, they built a functional prototype in just 25 hours. DeepSky detects potential flight hazards, such as drones, birds, and other objects on a collision course, offering crucial assistance in areas with limited or no air traffic control - especially at smaller airports and in uncontrolled airspace.
Unlike traditional systems, DeepSky is a simple and cost-effective solution. A single camera, costing only a few hundred PLN, is enough to significantly improve flight safety. Acting as an extra pair of "eyes" on board, the camera enhances a pilot’s situational awareness and minimizes collision risks.
DeepSky is a proactive step toward self-protection, offering a real opportunity to reduce risks in general aviation. This solution meets the needs of pilots, providing them with a tool to independently enhance safety in situations where air traffic control systems are insufficient.
3. What is DeepSky?
DeepSky is an innovative AI-powered application that enhances flight safety by monitoring airspace in real time. The system analyzes video feeds from onboard cameras, detecting potential hazards and alerting the pilot through visual notifications on the screen and voice messages, with a delay of only 3 to 5 seconds. Most of this time is spent processing the image using the GPT model, while generating the voice alert for the pilot takes less than 100 milliseconds, thanks to the “Eleven Flash v2.5” model.
The application leverages Voice AI technology and integrates with OpenAI, which supports data analysis and object recognition within the aircraft’s field of view. Notably, the AI not only identifies airborne objects but also intelligently ignores those on the ground, reducing unnecessary alerts. Additionally, the system can estimate the direction of detected objects, allowing pilots to react instantly and assess situations more effectively. Even the free version of ChatGPT can already detect distant aircraft with high accuracy.
DeepSky supports real-time video streaming from onboard cameras using Stream Video RTSP technology, significantly enhancing situational awareness and operational efficiency. Designed as a more affordable alternative to the TCAS system, DeepSky is also intended to function in offline mode, ensuring broader applicability in aviation. With this innovation, airspace can become safer, and AI technology more accessible to smaller aircraft.
4. How it works
Displaying the camera feed
The system starts by capturing and displaying a real-time video stream from the onboard camera. The pilot has direct access to an aerial situational overview via the interface screen.
Image analysis for obstacles
At regular intervals, the application captures a single video frame and sends it to the OpenAI (GPT-4o-mini) AI model along with a query to detect potential hazards. The model analyzes the image, identifies airborne objects (e.g., aircraft, drones, obstacles), and assesses collision risks.
Generating a voice alert
Based on OpenAI’s response, the application formulates a text-based warning, which is then sent to the ElevenLabs API. The Text-To-Speech technology generates a real-time voice message, providing the pilot with immediate information about detected threats.
Updating the radar view and playing the alert
The system visualizes detected obstacles on the screen as a radar-like situational map, helping the pilot quickly assess the situation. Simultaneously, a voice alert is played, allowing the pilot to react without taking their eyes off the cockpit.
This process runs continuously, analyzing the image, detecting threats, and delivering fast, intuitive information to the pilot in both visual and audio formats. By integrating AI and voice technology, DeepSky acts as an intelligent assistant, enhancing flight safety in dynamic airspace conditions.
5. Potential issues of the system
Despite the promising results, DeepSky faces challenges that need to be addressed before widespread implementation:
Required stable internet connection
Currently, DeepSky transmits images to OpenAI, which requires a fast internet connection - something that is not always available at high altitudes,Possible false alarms (false positives)While the OpenAI model is advanced, it is not specifically optimized for aviation and may generate incorrect recognitions.
6. How can we address these issues?
-
On-Edge AI (Edge computing)
Instead of relying on the cloud, the system could operate entirely locally, eliminating the need for internet access. This would make DeepSky independent of network conditions, Dedicated aviation neural network
Rather than using a general AI model, a specialized model could be developed, trained on images from aviation cameras, which would improve the accuracy of threat detection,Hardware optimization
Using efficient computational units (e.g. NVIDIA Jetson) would enable fast image processing with low power consumption.
7. What product can we create?
AI Alternative to TCAS
DeepSky can evolve into a full-fledged product:
- A complete AI-based collision detection system for General Aviation,
- Integrated hardware + software mounted in the pilot's cockpit,
- Cost significantly lower than traditional TCAS systems.
Instead of expensive radars, DeepSky will use advanced AI models and optical threat detection, reducing costs and opening a new era of technology accessibility for small aircraft pilots.
