Obstacle Avoidance System for UAVs using Computer Vision

Abstract

The purpose of this research is to develop an obstacle avoidance system for use on small, fixed-wing Uninhabited Aerial Vehicles (UAVs). In order to detect and avoid obstacles, computer based vision algorithms will be implemented with an automatic flight control system. Images of obstacles are captured using forward facing, externally mounted cameras. Obstacles will include moving and non-moving objects within the flight path of the UAV, which will be detected through the use of optical flow and feature-tracking methods. 1. Motivation and Goals for Research UAVs have the potential to replace inhabited aircraft for many civilian and military applications, which include, but not limited to, disaster relief assistance, search and rescue, and combat zone intelligence gathering. They have lower operating costs and pose minimal risk to human pilots. However, the lack of obstacle avoidance capabilities has limited the use of these vehicles. UAVs with inadequate sense and avoid capabilities are at risk of colliding with aerial objects, tall structures, and even trees. When collision takes place, damage to infrastructure and wildlife may occur, depending on the altitude and nature of the collision, in addition to the loss of the vehicle. Falling wreckage may pose even greater danger to building and humans underneath a UAV collision. For these reasons, current uses of UAVs are bound to military and restricted airspace. To be used in the National Airspace System (NAS), UAVs must comply with Federal Aviation Regulations requiring that they be equipped with obstacle avoidance systems. In the recent past, research has been performed regarding the use of quadrotor UAVs. The advantages of quadrotors include their maneuverability and their ability to hover. However, they have lower limits regarding their payload capacity, speed, operational ceiling, and range, when compared to fixed-wing airframes. The capabilities of fixed-wing UAVs provide users with the options to carry out long-range forest, desert, and maritime search-and-rescue missions and battlefield reconnaissance. Due to the high speed requirements of fixed-wing aircraft, sonar technology prevalent on quadrotor UAVs can no longer be used, for the reason that travel time of sonar waves decreases the range that the UAV can detect obstacles. The alternative is optical cameras, which do not depend on propagating waves, but instead use light as its medium to detect objects. This way, the UAV can instantaneously detect images and can avoid obstacles much sooner. Presently, Cal Poly Pomona is beginning development on an obstacle avoidance system for fixedwing UAVs. The intention is to create this system while using commercially available and low-cost components with the capacity for expansion. The rest of this paper will cover hardware selection and implementation, engineering and software approaches used, results thus far, and conclusion with future plans. 2. Hardware In order to maintain a relatively inexpensive obstacle avoidance system components were selected on the criteria of being reliable and cost effective. All components selected are commercially available off the shelf to ensure the above criteria. Table 2.1 lists all components of the system. Each component as follows were selected base cost and performance primarily. The Ardupilot is one of the most inexpensive fully functioning autopilots available and it performs as well as most autopilots far more expensive. The PX4 Pixhawk autopilot is also as cost effective as the Ardupilot and is marketed to be more powerful, but it is a new system and previous experience with the Ardupilot made it the best choice. The Arduino UNO was chosen due to its simplicity and cost, the processing power of the micro controller was not a factor due to the simple operations the micro controller would be preforming. The Super-Vision camera was chosen due to its small size, video quality and ability to adapt to light changes. The Panda board was chosen for its size, cost and previous experience using it for image processing. The Sig Kadet Senior air frame was selected for its payload capacity and stable characteristics. Finally the JR 12X was chosen for its reliability and range, in order to maintain the safest conditions while the plane is in flight. Figure 2.1 shows the system architecture and how information will flow through the avoidance system.