Unmanned aerial vehicle avoidance algorithm based on improved ssd target detection network

Abstract

The invention discloses an unmanned aerial vehicle avoidance algorithm based on an improved ssd target detection network. The method comprises the steps of 1, making a binocular view angle image dataset of an obstacle; 2, establishing a target detection model according to the calibration data set; 3, constructing an ssd target detection network, and training the network parameters by using the data set; 4, transmitting each pair of obstacle pictures in the test set to a trained ssd detection network, and outputting category labels and position labels predicted by the neural network; 5, utilizing different visual angles to detect the position of the obstacle, and performing binocular distance measurement to obtain an obstacle distance; and 6, feeding back the obstacle classification and distance to an unmanned aerial vehicle driving system to control the unmanned aerial vehicle driving system to automatically avoid. According to the method, the ssd target detection algorithm and the binocular distance measurement image algorithm are combined, so that the combined detection result can effectively obtain the type of the obstacle, and the vector distance between the obstacle and the unmanned aerial vehicle in the world coordinate position can be accurately calculated.

Description

[0001] Technical field: [0002] The invention relates to the field of unmanned aerial vehicle avoidance, in particular to an unmanned aerial vehicle avoidance algorithm based on an improved ssd target detection network. [0003] Background technique: [0004] With the rapid development of science and technology, UAVs have been widely used in the military and civilian fields, but the increase in the amount of airspace flight activities followed, and the probability of accidents also increased. This will not only cause economic loss, but even have a serious impact on flight missions. To ensure that UAVs can be safe and reliable when performing various tasks, it is inseparable from the assistance of navigation and avoidance systems. [0005] In 2015, DJI launched the Guidance Perceptual Obstacle Avoidance System. As an industry leader, DJI’s Perceptual Obstacle Avoidance System is a composite multi-directional obstacle avoidance system that supports avoidance in 5 directions at ...

AI Technical Summary

Problems solved by technology

[0005] In 2015, DJI launched the Guidance Perceptual Obstacle Avoidance System. As an industry leader, DJI’s Perceptual Obstacle Avoidance System is a composite multi-directional obstacle avoidance system that supports avoidance in 5 directions at the same time. The Matrice100 launched in the second year was built on the This system, but the avoidance direction of the system is not comprehensive, and the avoidance algorithm is relatively mechanical. From the perspective of the overall market application effect, the research on the perception and avoidance system has not been perfected. , the market demands and provides for

[0006] Due to the limited detection range and accuracy of the sensor, it is not yet possible to achieve true all-round obstacle avoidance. There are blind spots, and the obstacle avoidance performance has high requirements on the environment. Obtaining sufficient scene attributes with a technical level to provide effective guidance for UAVs to avoid obstacles has become the top priority for the large-scale application of UAVs in my country.

Images