Development of Intelligent Human Flow Density Detection System based on Sensor Fusion

Crowd counting and human flow estimation is the critical technique for modern sociality safety. In this study, the intelligent human flow density detection system based on sensor fusion technique are developed. The system includes three kinds of subsystems. The image subsystem employs SSD neural network and CSRNet to estimate human counting and crowd density map. The LiDAR subsystem uses Voxel Grid filter, RANSAC estimation, KD tree background subtraction and Kalman filter to detect and track human flow. Through using Image and LiDAR chessboard calibration algorithm, the extrinsic calibration parameters of the 3D-2D transformation matrix are estimated. The system can re-project distance information of point clouds to the image space and display the human density map in point cloud space. Three kinds of different crowd environmental conditions were selected to verify the performance of the proposed system. The results of experiment testify the proposed system achieved human counting and human flow detection in the scan field.