Power and Telecommunication Lines Detection and Avoidance for Drones

Power and telecommunication lines are hard to detect using sensors such as ultrasonic sensor, LIDAR etc. These sensors are also expensive and uses some amount of power which reduce flight time. This has made it hard for drone application fields such as delivery to successfully delivery packages to their destination especially in rural areas with a lot of these lines or cables around the streets. This problem has resulted in the ban of drone deliveries in most areas since it isn't considered to be safe [1]. Our goal is to use computer vision with deep learning to detect power and telecommunication lines. This will limit the number sensors and henceforth reduce the overall production cost of delivery drones. This study will also aim to reduce the risk or probability associated with a drone crashing into power and telecommunication lines which can result in a loss or injury to nearby people. We use the tensorflow, Keras and OpenCV python library to train and test the model for this research with 1300 image datasets. This approach only works during the day with presence of light. Aside from this downside, the model has performed well in detecting cables given complex situations such as angle, cable thickness, color and length.