SPYDR

Unmanned aerial vehicles (UAVs) are often capable of performing tasks that are dangerous or costlier for humans. Examples include monitoring of tall structures or large areas, surveillance of a complex terrain, and search and rescue missions. We will use a quadcopter to develop a system for monitori

Project Title

SPYDR

Project Area of Specialization

Artificial Intelligence

Project Summary

Unmanned aerial vehicles (UAVs) are often capable of performing tasks that are dangerous or costlier for humans. Examples include monitoring of tall structures or large areas, surveillance of a complex terrain, and search and rescue missions. We will use a quadcopter to develop a system for monitoring and surveillance of a designated area of interest. The project will involve writing a mission software and enabling auto-pilot to fly the UAV. This software will be capable of loading a mission, and fetching and displaying live flight information and video. We will also perform real-time object (and threat) detection using computer vision from the relayed video feed. One of objectives of the threat detection will be to raise alarms if an unexpected or malicious object is encountered during flight.

Project Objectives

• This software will provide a map interface that will enable the user to mark way points the cordinates of which will be fed to the quad copter. 
• The quadcopter will fly on auto mode on the planned trajectory using the cordinates. 
• This software will be capable of streaming live flight video from the quad-copter’s camera, on a GUI interface.
• The feed will be processed for real-time object classification (human beings or specifically intruders) using computer vision from the relayed video feed.
• It will detect the classified object’s presence/location inside the video feed by drawing a bounding box around it.
• An alarm will be generated whenever an intruder is detected inside the stream.

Project Implementation Method

1. Intruder or threat detection is a classic object detection and classification problem. The live stream coming from the drone is divided into frames which are processed to detect and classify intruders or threats. So the question is, what classifies as an intruder/threat? Since the project is in its early development phase, a regular pedestrian classifies as one at the moment but the sky is the limit. Vehicles and weapons are an additional aim my project team is hoping to achieve. We are employing Tensorflow's Object detection API and Single Shot detector as a model. It is approximately 74 to 76% accurate with an exceptional real-time performance. With transfer learning using our own data collected in the university premises, we will further train SSD v1 mobilenet which is pre-trained on COCO dataset. The stream received will be divided into frames using python's open cv and the model will individually process each frame for detecting and classifying intruders by drawing bounding boxes around them which will then trigger an alarm.

2. 3DR SOLO's compatibility with dronekit python serves to accomplish mission planning or trajectory planning. API provides classes and methods to connect to a vehicle through a script, get or set vehicle telemetry information, create and manage way-point missions and much more. Using a suitable map, the software will make the user capable of drawing a path, the coordinates of which will be fed to the python script that will communicate them to the SOLO. Altitude and air speed will be under consideration as well. This SOLO will follow the mission given to it until the session expires.

Benefits of the Project

Drones for surveillance is the future of CCTV. Mobility, range and altitude are the key attributes that make drones such as quad copters stand out and encourage their use for monitoring and surveillance of areas where circuit cameras can't reach or human eye fails to perceive. Considering the heightened concerns regarding scrutiny of critical infrastructure, borders or hostile demonstration environments, drones can keep a close eye for intrusions. Combined with AI, intruders can be detected from the live stream received from the drone automatically, making it highly effective for security systems.

Technical Details of Final Deliverable

The idea is to build a software that plans a trajectory using the map of area of interest and obtaining the way-points. With the aid of Dronekit-python API's way-point mission creation and management over MAVLINK protocol with the SOLO, the need for a controller to fly it is eliminated. Paired up with Tensorflow's object detection API, intruders in the live stream from the Go Pro Hero 4 attached to the SOLO can be detected on which the system generates an alarm.

Final Deliverable of the Project

Software System

Core Industry

IT

Other Industries

Security

Core Technology

Artificial Intelligence(AI)

Other Technologies

Robotics

Sustainable Development Goals

Industry, Innovation and Infrastructure

Required Resources

If you need this project, please contact me on contact@adikhanofficial.com