Surveillance LiDAR fast detection of moving targets based on CFAR

When LiDAR works in the long-range surveillance mode, the target point cloud is relatively sparse. Detecting and identifying moving targets by deep learning method requires high hardware conditions and has low real-time performance and low recognition rate. The traditional moving target detection method of two-dimensional video images has high false alarm rate, while the three-dimensional Gaussian mixture model method has low false alarm rate but low processing speed. In view of the need of rapid detection of moving targets by LiDAR in practical projects and the fact that echo intensity of the lidar obeys the log-normal distribution, method of processing point cloud data by using the Constant False Alarm Rate (CFAR) detection method of clutter map is proposed. The false alarm rate and detection probability are analyzed, and a comparative experiment is made with other methods based on change detection. Experiments show that when using Log-t clutter map CFAR method to detect cooperative targets, the false alarm rate can be close to zero when the missing alarm rate is zero, which can be controlled to reach a very low false alarm rate, and the detection time is close to 1/18 of the three-dimensional Gaussian mixed model method, meeting the practical engineering requirements of high real-time performance and low false alarm rate.