With the widespread use of UAVs in daily life, there are many sensors and algorithms used to ensure flight safety. Among these sensors, lidar has been gradually applied to UAVs due to its stability and portability. However, in the actual application, lidar changes its position with the movement of the UAV, resulting in an offset in the detected point cloud. What's more, when the lidar works, it scatters laser light from the center to the surroundings, which causes the detected point cloud to be externally sparse and dense inside. This point cloud with uneven density is difficult to cluster using common clustering algorithms. In this paper, a velocity estimation method based on the polynomial fit is used to estimate the position of the lidar as it scans each point and then corrects the twisted point cloud. Besides, the clustering algorithm based on relative distance and density (CBRDD) is used to cluster the point cloud with uneven density. To prove the effectiveness of the obstacle detection method, the simulation experiment and actual experiment were carried out. The results show that the method has a good effect on obstacle detection.

Real-time and stable positioning data is essential for the UAV to perform various tasks. The traditional multi-sensor data fusion algorithm needs to know the measurement noise of sensor data, and even if there are corresponding adaptive methods to estimate the noise, most methods cannot deal with time-varying noise. In addition, traditional fusion algorithms usually are complicated, causing a large amount of calculation. In this paper, a multi-sliding window classification adaptive unscented Kalman filter (MWCAUKF) method with timestamp sort updating was proposed, which can improve the accuracy and stability of positioning. This method consists of three phases. First, according to the timestamp of sensor data, the multi-sensor data are added with fusion filtering in order. Then it estimates the measurement noise of multiple sensors through multiple sliding Windows. Finally, the sensor data classification method is adopted to deal with the filter instability caused by time-varying noise. Both theoretical analysis and experimental results show that this method has a low computational cost, high accuracy, and good stability. INDEX TERMS Multi-sensor fusion, unmanned aerial vehicle, positioning, adaptive Kalman filtering.

No abstract

No abstract

No abstract