Jianfeng Sun

3D pose estimation of ground rigid target based on ladar range image.

In the target recognition of laser radar (ladar), the accurate estimation of target pose can effectively simplify the recognition process. To achieve 3D pose estimation of rigid objects on the ground and simplify the complexity of the algorithm, a novel pose estimation method is proposed in this paper. In this approach, based on the feature that most rigid objects on the ground have large planar areas which are horizontal on the top of the targets and vertical sides and combined with the 3D geometric characteristics of ladar range images, the planar normals of rigid targets were adopted as the vectors in the positive direction of the axes in the model coordinate system to estimate the 3D pose angles of targets. The simulation experiments were performed with six military vehicle models and the performance in self-occlusion, occlusion, and noise was investigated. The results show that the estimation errors are less than 2° in self-occlusion. For the tank LECRERC model, as long as the upper and side planes of the target are not completely occluded, even though the occlusion reaches 80%, the pose angles can be estimated with the estimation error less than 2.5°. Moreover, the proposed method is robust to noise and effective.

Research of underwater target detection using a Slit Streak Tube Imaging Lidar

Slit Streak Tube Imaging Lidar (STIL) is a promising imaging system as its high frame rate and good image quality. As its important applications, the depth under the water to detect a underwater target near the coast is less than 20m. It can effectively indemnify the safety of ships. We use a Slit Streak Tube Imaging Lidar, including a laser whose wavelength is 532nm and the energy of one pulse is 20mJ. Through the outfield experiment on the sea, we discussed the optical properties of the target surface and distinguished two targets with the distance more than 10cm in one image. Also we get the imaging of the target which is 6m deep in the sea. We calculated the relationship between lidar detectable range and laser power in different attenuation. It will provide the basis for the future experiments.

Model-based recognition of 3D articulated target using ladar range data.

Ladar is suitable for 3D target recognition because ladar range images can provide rich 3D geometric surface information of targets. In this paper, we propose a part-based 3D model matching technique to recognize articulated ground military vehicles in ladar range images. The key of this approach is to solve the decomposition and pose estimation of articulated parts of targets. The articulated components were decomposed into isolate parts based on 3D geometric properties of targets, such as surface point normals, data histogram distribution, and data distance relationships. The corresponding poses of these separate parts were estimated through the linear characteristics of barrels. According to these pose parameters, all parts of the target were roughly aligned to 3D point cloud models in a library and fine matching was finally performed to accomplish 3D articulated target recognition. The recognition performance was evaluated with 1728 ladar range images of eight different articulated military vehicles with various part types and orientations. Experimental results demonstrated that the proposed approach achieved a high recognition rate.

Correlation target recognition for laser radar

The target recognition of laser radar is a hot research because laser radar can produce the intensity and range imagery. Laser radar has high space resolution, and can obtain rich target information. Correlation recognition has been used to many fields, such as infrared as well as synthetic aperture radar (SAR). In this paper, the two filters are used in experiment of laser radar. MACH filter is used to detect the target, and DCCFs are used to recognize the unknown target. The samples are generated by OpenGL technology, and the filters are designed using the simulated ladar images. The test samples are added noise according to the imaging principle of laser radar. Two sample sets, one adding noise, another filtering the noise, are used in order to contrast the different performance. At last, the experiment results are given.

Object recognition of ladar with support vector machine

Intensity, range and Doppler images can be obtained by using laser radar. Laser radar can detect much more object information than other detecting sensor, such as passive infrared imaging and synthetic aperture radar (SAR), so it is well suited as the sensor of object recognition. Traditional method of laser radar object recognition is extracting target features, which can be influenced by noise. In this paper, a laser radar recognition method-Support Vector Machine is introduced. Support Vector Machine (SVM) is a new hotspot of recognition research after neural network. It has well performance on digital written and face recognition. Two series experiments about SVM designed for preprocessing and non-preprocessing samples are performed by real laser radar images, and the experiments results are compared.

Registration of partially overlapping laser-radar range images

Abstract. To register partially overlapping three-dimensional point sets from different viewpoints, it is necessary to remove spurious corresponding point pairs that are not located in overlapping regions. Most variants of the iterative closest point (ICP) algorithm require users to manually select the rejection parameters for discarding spurious point pairs between the registering views. This requirement often results in unreliable and inaccurate registration. To overcome this problem, we present an improved ICP algorithm that can automatically determine the rejection percentage to reliably and accurately align partially overlapping laser-radar (ladar) range images. The similarity of k neighboring features of each nonplanar point is employed to determine reasonable point pairs in nonplanar regions, and the distance measurement method is used to find reasonable point pairs in planar regions. The rejection percentage can be obtained from these two sets of reasonable pairs. The performance of our algorithm is compared with that of five other algorithms using various models with low and high curvatures. The experimental results show that our algorithm is more accurate and robust than the other algorithms.

Target recognition of laser radar using correlation filter with in-plane rotation invariance

Laser radar can simultaneously produce the range image and the intensity image, and it can directly collect rich information of target. Compared with the other sensors, such as infrared or radar, laser radar can enhance the recognition rate and the precision of target aimed point. When laser radar vertically detects the objects on the plane ground, the correlation filters with in-plane rotation invariance are usually used to solving the problem of the target recognition. Traditional correlation filters are still improved on the aspect of recognition rate. In the paper, through deducing the relationship between support vector machine (SVM) and correlation principle in the signal processing, a new correlation filter, named linear SVM correlation filter (LSCF) that has the properties of SVM, is proposed. The real images of laser radar are used as the training and testing samples. The experiments state that the filter has good recognition attributes, such as stable correlation output and high recognition rate. LSCF is suitable to be the recognition algorithm of the imaging laser radar.

A fast image seeking algorithm based on imaging ladar

The quantum genetic algorithm is a fast algorithm for searching global optimal solution of a function, being proposed in recent years. The algorithm introduces the method of quantum, which makes the algorithm has high parallelism in solving problems. So the algorithm can be used to the quick searching. In this paper, the algorithm is introduced to target seeking of the ladar imagery. Through theoretical analysis and repeated experiments, the suitable self-adapting quantum revolving door and the effective preprocess of the ladar imagery have been found. This paper also makes use of the method of quantum transition and improves the accuracy of the algorithm by fusing the range image with the intensity image. Moreover, the preliminary quantum genetic algorithm for imaging ladar has been established. To some extent, with the use of this algorithm the best matching points can be found fast and accurately.

Correlation detection filter for imaging laser radar

Laser radar can simultaneously produce the intensity and range images, and the space resolution is high, so the recognition performance is well, and it can choose the aim point of target. Laser radar is applied to many fields, such as guidance, navigation, and becomes the research hot point in recent years. In the vertical detection of laser radar, the algorithm is required not only solving in-plane rotation-invariant problem, also the distortion-invariant problem, and it must satisfied the real-time. Correlation algorithm is a parallel processing procedure, detecting many targets at one time, and its design can be implemented on the high speed digital signal processor. In the paper, a new filter named CHF-MACH filter is presented, which combine multiple circular harmonic expansions into one filter through MACH criteria. Because of the filter having the characters of the two filters, it can solve the problems of in-plane rotation-invariance and distortion-invariance simultaneously, and meet the real-time requirement. The simulated range image of laser radar is regarded as research target, and computing the PSR (peak to sidelobe ratio) values of correlation output of the different objects, and plotting the PSR curves of the different angles. Simulating the scene of laser radar which includes multiple objects, CHF-MACH filter performance is validated through testing with the different angles for the objects, and the non-training images can obtain the well correlation output.