CN-122017845-A - Obstacle detection method and device based on multi-millimeter wave radar data

CN122017845ACN 122017845 ACN122017845 ACN 122017845ACN-122017845-A

Abstract

The application discloses an obstacle detection method and system based on multi-millimeter wave radar data, and relates to obstacle detection, wherein the obstacle detection method comprises the steps of collecting multi-path millimeter wave radar data; the method comprises the steps of carrying out FFT conversion on each path of radar data, extracting radar point cloud data, marking radar ID and time stamp of each point cloud data to obtain a multi-source point cloud data set with identification information, carrying out coordinate conversion on the multi-source point cloud data set with the identification information according to the installation position and angle of each millimeter wave radar under a vehicle body coordinate system to obtain a spatially registered point cloud data set, carrying out fusion processing on redundant point cloud data generated by repeated detection of multiple radars on the same obstacle to obtain fusion point cloud data after redundancy elimination, matching the fusion point cloud data at continuous moments by adopting a data association algorithm, and carrying out prediction and update on the motion state of the obstacle by utilizing a filtering algorithm to obtain a dynamic target tracking result comprising the position and the speed of the obstacle.

Inventors

HUANG YAN
LI SHUANGHUA
FENG CHONG
CHEN ZHIYONG

Assignees

理工雷科智途(北京)科技有限公司
理工雷科智途(泰安)汽车科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260225

Claims (10)

1. An obstacle detection method based on multi-millimeter wave radar data is characterized by comprising the following steps: S1, acquiring multiple paths of millimeter wave radar data in parallel to obtain radar data streams containing original echo data; S2, performing FFT (fast Fourier transform) on each path of radar data, extracting radar point cloud data, and marking a radar ID and a time stamp to which each point cloud data belongs to obtain a multi-source point cloud data set with identification information; S3, carrying out coordinate transformation on the multi-source point cloud data set with the identification information according to the installation position and the angle of each millimeter wave radar under the vehicle body coordinate system, and converting the point cloud data under each radar local coordinate system into the vehicle body coordinate system to obtain a point cloud data set with spatial registration; s4, identifying point cloud data in an adjacent radar overlapping detection area according to the radar ID and the three-dimensional coordinates of the point cloud in a vehicle body coordinate system, and carrying out fusion processing on redundant point cloud data generated by repeated detection of a plurality of radars on the same obstacle to obtain fusion point cloud data after redundancy elimination; s5, matching the fusion point cloud data at continuous moments by adopting a data association algorithm according to the time stamp, establishing an obstacle target track, and predicting and updating the motion state of the obstacle by utilizing a filtering algorithm to obtain a dynamic target tracking result comprising the position and the speed of the obstacle; And S6, mapping the obstacle information into a preset grid map according to the redundancy-removed fusion point cloud data and the dynamic target tracking result, and calculating the occupation probability of each grid through a probability updating rule to obtain a real-time updated 360-degree obstacle grid map.
2. The obstacle detection method based on multi-millimeter wave radar data according to claim 1, comprising: s1, acquiring a plurality of millimeter radar wave data in parallel, including: 8 paths of millimeter radar wave data are synchronously acquired through 4 millimeter wave angle radars arranged at the front, the rear, the left and the right of the vehicle and 4 millimeter wave angle radars arranged at the front, the rear, the left and the rear of the left; Wherein, the detection areas of adjacent millimeter wave angle radars are provided with an overlapping angle of 15 degrees to 30 degrees, so that the detection ranges of 8 millimeter wave radars jointly cover the 360-degree range around the vehicle; millimeter wave angle radar adopts 77GHz or 79GHz frequency band.
3. The obstacle detection method based on multi-millimeter wave radar data according to claim 2, comprising: s2, obtaining a multi-source point cloud data set with identification information, wherein the multi-source point cloud data set comprises: performing adaptive threshold detection on millimeter wave radar data to obtain denoised radar data; Performing FFT (fast Fourier transform) on the denoised radar data to obtain the amplitude and the phase of N frequency components, wherein N is the number of FFT points; According to the radar equation Converting the frequency component into corresponding distance information, wherein f is frequency, v is electromagnetic wave speed, r is target distance, Obtaining distance distribution data of the obstacle targets as wavelengths; According to the millimeter wave angle radar, a plurality of built-in receiving antennas are utilized to acquire multichannel receiving signals of the same target, the phase information of distance points in each receiving antenna channel is extracted for the distance points with the amplitude exceeding a preset threshold value, and the phase difference of adjacent antenna channels is calculated And according to Calculating a target azimuth angle theta, wherein d is the distance between adjacent antennas; Combining the distance r, azimuth angle θ and amplitude A of each obstacle target to form point cloud data ; Adding a radar ID number for generating the data and a time stamp t for data acquisition to each point cloud data to obtain a format of Is provided with identification information.
4. The obstacle detection method based on multi-millimeter wave radar data according to claim 3, comprising: performing adaptive threshold detection on millimeter wave radar data to obtain denoised radar data, including: calculating the amplitude mean mu and standard deviation sigma of the echo data of the current frame; Setting a detection threshold as mu+ksigma, wherein k is a constant of 3-5; And setting echo data with the amplitude lower than a detection threshold to zero to obtain denoised radar data so as to filter weak reflection clutter generated by underground water mist and dust.
5. The obstacle detection method based on multi-millimeter wave radar data according to claim 3, comprising: s3, obtaining a spatially registered point cloud data set, which comprises the following steps: the installation parameters of each millimeter wave radar under the vehicle body coordinate system are obtained, including installation position coordinates And angle of installation , wherein, As a translation vector of the radar with respect to the origin of the car body coordinate system, The rotation angles of the radar around the axis of the vehicle body coordinate system X, Y, Z are respectively; Searching installation parameters of corresponding radars according to the radar ID of each point cloud data in the multi-source point cloud data set with the identification information; Point cloud data Polar coordinates of (a) To cartesian coordinates in a radar local coordinate system, wherein, , , ; According to the installation angle Calculating a rotation matrix R from a radar local coordinate system to a vehicle body coordinate system; Through a coordinate transformation formula Converting the point cloud coordinates under the radar local coordinate system into a vehicle body coordinate system; preserving the information of the amplitude A, the radar ID and the time stamp t of the point cloud to obtain the format of the vehicle body coordinate system Is provided for the spatial registration of the point cloud dataset.
6. The obstacle detection method based on multi-millimeter wave radar data according to claim 5, comprising: s4, obtaining fusion point cloud data subjected to redundancy removal, wherein the fusion point cloud data comprises the following steps: determining the boundary of the overlapping detection area of each adjacent radar pair according to the overlapping angle of the detection areas of the adjacent millimeter wave angle radars; Registering each point cloud data in a point cloud data set according to space By means of coordinates Judging whether the corresponding point cloud is positioned in the overlapped detection area; Calculating the distance between point clouds of different radar IDs according to the point cloud data in the overlapping detection area ; Distance is to Less than a preset threshold Is determined as a repeated detection point of the same obstacle, wherein, Setting according to the distance resolution of the millimeter wave radar; Fusing a plurality of repeated detection points determined to be the same obstacle; And reserving the maximum amplitude value and the earliest timestamp of the fused point cloud, deleting the repeated point cloud, and obtaining the fused point cloud data after redundancy elimination.
7. The obstacle detection method based on multi-millimeter wave radar data according to any one of claims 2 to 6, comprising: s5, obtaining a dynamic target tracking result comprising the position and the speed of the obstacle, wherein the dynamic target tracking result comprises the following steps: dividing fusion point cloud data into continuous data frames according to time stamps t in the fusion point cloud data after redundancy removal; for each point cloud in the current frame, according to the corresponding coordinates Searching candidate matching points with the distance smaller than an association threshold in the point cloud data of the previous frame, wherein the association threshold is calculated according to the time interval of the adjacent frames and the maximum running speed of the vehicle; Adopting a nearest neighbor data association algorithm to associate the point cloud of the current frame with the point closest to the candidate matching point of the previous frame to obtain an associated point cloud sequence so as to establish a corresponding relation of the cross-frame point clouds; constructing barrier target tracks according to the successfully-associated point cloud sequences, wherein each track comprises a barrier target ID, a historical position sequence and a time stamp sequence; performing state estimation on each obstacle target track by adopting a Kalman filtering algorithm, wherein a state vector comprises an obstacle target position (x, y, z) and a speed Estimating the current time position according to the previous time state through a prediction step, and correcting a prediction result through an update step by utilizing the current observation value; Outputting each tracked obstacle target ID, position (x, y, z) and velocity And obtaining a dynamic target tracking result comprising the position and the speed of the obstacle.
8. The obstacle detection method based on multi-millimeter wave radar data according to claim 7, comprising: Obtaining a dynamic target tracking result including the position and the speed of the obstacle, and further comprising: and for the continuous multi-frame non-updated target tracks, judging that the target disappears and deleting the corresponding tracks.
9. The obstacle detection method based on multi-millimeter wave radar data according to claim 7, comprising: s6, obtaining a real-time updated 360-degree obstacle grid map, which comprises the following steps: Creating a grid map with an origin of a vehicle body coordinate system as a center, and setting the grid resolution so that a map range covers a 360-degree area around the vehicle; coordinates of each point in the redundancy-removed fusion point cloud data Mapping to a corresponding grid index in the grid map; calculating the grid occupation probability increment according to the point cloud data mapped to each grid; Updating the occupancy probability of each grid by adopting a probability updating rule and combining the historical occupancy probability of the grid with the current observation data; Marking the grid as an obstacle grid, a free grid or an unknown grid according to the occupancy probability and a preset threshold; A 360 ° obstacle grid map is output that contains grid occupancy states and occupancy probabilities.
10. An obstacle detection device based on multi-millimeter wave radar data, comprising: the data acquisition module is used for acquiring multiple paths of millimeter wave radar data in parallel; The point cloud extraction module is used for carrying out FFT (fast Fourier transform) on each path of radar data, extracting radar point cloud data, and marking a radar ID (identity) and a time stamp to which each point cloud data belongs to obtain a multi-source point cloud data set with identification information; the coordinate transformation module is used for carrying out coordinate transformation on the multi-source point cloud data set with the identification information according to the installation position and the angle of each millimeter wave radar under the vehicle body coordinate system, and converting the point cloud data under each radar local coordinate system into the vehicle body coordinate system to obtain a point cloud data set with spatial registration; The data fusion module is used for identifying point cloud data in an adjacent radar overlapping detection area according to the radar ID and the three-dimensional coordinates of the point cloud in a vehicle body coordinate system, and carrying out fusion processing on redundant point cloud data generated by repeated detection of a plurality of radars on the same obstacle to obtain fusion point cloud data after redundancy elimination; The target tracking module is used for matching the fusion point clouds at continuous moments by adopting a data association algorithm according to the time stamp, establishing an obstacle target track, and predicting and updating the motion state of the obstacle by utilizing a filtering algorithm to obtain a dynamic target tracking result comprising the position and the speed of the obstacle; And the grid map module maps the obstacle information into a preset grid map according to the redundancy-removed fusion point cloud data and the dynamic target tracking result, calculates the occupancy probability of each grid through a probability updating rule, and obtains a real-time updated 360-degree obstacle grid map.

Description

Obstacle detection method and device based on multi-millimeter wave radar data Technical Field The application relates to the field of obstacle detection, in particular to an obstacle detection method and device based on multi-millimeter wave radar data. Background Along with the promotion of intelligent mine construction, the automatic driving technology of the underground transportation equipment becomes one of key technologies for improving the safe production and transportation efficiency of the coal mine. However, the well and mining specific operating environment presents serious challenges to the vehicle perception system. The underground roadway is narrow, light is dim, a large amount of water mist and dust exist, and the application effect of the traditional perception technology is severely restricted by the severe environmental factors. The existing perception scheme of the underground transportation vehicle mainly adopts a multi-sensor fusion framework with a laser radar as a main component and a camera and a millimeter wave radar as an auxiliary component. Specifically, the main laser radar is usually arranged at the top of the vehicle and is responsible for sensing the environment of the whole body of the vehicle, the camera is arranged right in front of the vehicle and is matched with the laser radar to realize the recognition of the obstacle in front, the millimeter wave radar is arranged at the left front and the right front of the vehicle and is used as the supplement of the laser radar to detect whether the two sides of the vehicle can safely pass through. These sensor data are processed by means of point cloud scale, either fusion-by-fusion (pre-fusion) or target-level fusion (post-fusion). However, the conventional sensing schemes described above have a number of limitations in the well and mining environment, in which, first, lidar faces serious performance degradation problems in the downhole environment. The water mist discharged from the underground chamber in the underground tunnel can cause laser scattering to cause a large number of false detection, the dust environment can absorb and scatter laser signals to obviously reduce the detection distance and accuracy, and meanwhile, the laser radar has high cost, and the generated mass point cloud data needs high-performance computing resources for real-time processing. Second, the camera is almost out of order in a dark downhole environment. Even if the light supplementing device is equipped, the perception performance of the camera is extremely unstable under the complex illumination condition of strong light and weak light alternation. The deep learning algorithm based on vision has the problems of false detection and omission, a large amount of annotation data is required for training, and the model decision process lacks of interpretability. Again, although millimeter wave radars have the natural advantage of penetrating water mist and dust, and still can maintain stable detection performance in severe environments, the detection angle of a single millimeter wave radar is usually only ±60°, and the requirement of 360 ° omnidirectional obstacle detection around the vehicle cannot be met. In addition, the multi-sensor data fusion also faces technical challenges such as data redundancy and conflict, high algorithm complexity, strict real-time requirements and the like. Noise characteristics and uncertainty differences of different sensors increase the design difficulty of the fusion algorithm. Therefore, how to fully exert the anti-interference advantage of the millimeter wave radar in the underground severe environment, break through the physical limitation of the detection angle of the single radar through the cooperative work of a plurality of millimeter wave radars, realize the 360-degree high-reliability obstacle detection of the whole vehicle in the underground narrow roadway environment, and become the technical problem to be solved urgently. Disclosure of Invention Aiming at the problems that the traditional laser radar and the camera are invalid in perception due to water mist, dust and darkness in the underground narrow roadway environment and the vehicle cannot be detected in all directions due to limited detection angle of a single millimeter wave radar, the application provides the obstacle detection method and device based on multi-millimeter wave radar data, the limitation of the physical detection angle of the single radar is broken through the cooperative arrangement and data fusion processing of 8 millimeter wave angle radars, and the 360-degree real-time obstacle detection of the whole vehicle in the underground narrow roadway environment is realized. The obstacle detection method based on the multi-millimeter wave radar data comprises the steps of S1, collecting multi-millimeter wave radar data in parallel, S2, carrying out FFT (fast Fourier transform) on each path of radar data, extracting radar point cloud data,