With the large-scale application of AI technology in the field of unmanned aerial vehicles, UAVs began to become more intelligent.
Not only can do real-time tracking lock to take the real-time processing target information can also be done automatically identify avoid obstacles.
Behind these actions is to break the UAV computer vision technology.
Computer vision technology, is simply a sensor camera + computer simulation similar to the human eye and brain function, to perceive a three-dimensional space around, and further identify objects, and other motion state is determined.
In the UAV field of computer vision technology to solve two major problems. One is the perception of distance, a target is detected.
Distance perception, that perception of the surrounding environment in real-time, automatic identification of the main solution is to avoid the barriers. Air environment, although not as complex ground environment, but will also face many unknown risks, such as birds, such as some high objects. UAVs in civilian areas where, due to the impact of obstacles lead to damage of the UAV occupy most of the damage in all cases.
Object detection, including the detection, tracking and recognition, navigation and the like. Analysis can identify ground targets, real-time tracking shot is a good example of the application.
In fact, the perception of distance and target detection, can be summarized as "real-time monitoring and analysis of the surrounding scene." This requires not only UAVs have excellent information gathering capabilities, UAVs also need to have good information processing judgment.
Information-gathering capabilities depends on the hardware, such as radar, cameras, etc., and information processing judgment depends on the algorithm.
It is well known in the field of artificial intelligence, algorithms model the effect of feeding it depends on the quality of the data, the UAV field as well. In order to give drones the ability to handle more complex scenarios, pre-fed the data necessary to achieve massive, fine, scene of, authenticity, and these data are all derived from the data label.
UAV art mainly applied to the following types of data are denoted:
1.2D box
Mark the vehicle:
2. Polygon
Mark the vehicles and people:
3. Semantic Segmentation
Different areas of the picture is divided marked:
4. Video Labels
Video tracking label vehicles and pedestrians
In these real blessing data, algorithms UAV model effect will be more precise, to the scene to determine the flight will be more reasonable, UAVs can play a greater role in the following areas:
1. Vehicle Recognition and Tracking
After the drones with cameras, based on the depth of learning, for real-time identification and analysis of the characters and vehicles on the road.
2.-ground probe
Drones speed motion cameras and radar can be effectively ground and imaging 3D map rendering, thereby providing a 3D perspective view and 2D floor plan view of relevant institutions.
3. UAV delivery
With laser radar and other sensors, UAVs can automatically avoid obstacles in the course of the flight, analyze and identify ground targets, to find the most appropriate courier delivery location.
Future applications of UAVs will not only limited to this, with the gradual development of data annotation and computer vision technology, the application of UAVs will create more possibilities.