AI Assisted Pedestrian IDENTIFICATION Using DRONES
Abstract (300 words maximum)
This paper examines research on pedestrian identification and enumeration using drone based object detection frameworks. Potential applications include crowd counting at outdoor events, search and rescue operations to locate missing people, and security monitoring in large open spaces. Our team implemented open-source and enhanced AI models to enable real-time pedestrian detection directly on a computer interface. Using the Skydio X10 drone, we captured original footage to evaluate each model’s effectiveness in identifying humans from varying altitudes and varying light clarity. The results indicate that YOLOv12x and YOLOv10x were the most effective models at detecting pedestrians. Our additions to the existing models further enhanced the detectability, and enumeration was added to the model. Testing with footage from the Skydio X10 displayed that these models maintained consistent and strong performance across a wide range of altitudes. However, detection accuracy declined significantly once the drone exceeded 70 feet in altitude. These findings suggest that even though there are numerous practical applications for pedestrian detection under 70 feet, extending reliable detection capabilities beyond this altitude would require more advanced hardware and higher quality imaging systems.
Use of AI Disclaimer
no
Academic department under which the project should be listed
SPCEET – Industrial and Systems Engineering
Primary Investigator (PI) Name
Adeel Khalid
AI Assisted Pedestrian IDENTIFICATION Using DRONES
This paper examines research on pedestrian identification and enumeration using drone based object detection frameworks. Potential applications include crowd counting at outdoor events, search and rescue operations to locate missing people, and security monitoring in large open spaces. Our team implemented open-source and enhanced AI models to enable real-time pedestrian detection directly on a computer interface. Using the Skydio X10 drone, we captured original footage to evaluate each model’s effectiveness in identifying humans from varying altitudes and varying light clarity. The results indicate that YOLOv12x and YOLOv10x were the most effective models at detecting pedestrians. Our additions to the existing models further enhanced the detectability, and enumeration was added to the model. Testing with footage from the Skydio X10 displayed that these models maintained consistent and strong performance across a wide range of altitudes. However, detection accuracy declined significantly once the drone exceeded 70 feet in altitude. These findings suggest that even though there are numerous practical applications for pedestrian detection under 70 feet, extending reliable detection capabilities beyond this altitude would require more advanced hardware and higher quality imaging systems.