In this paper we present a novel way to analyze LADAR images and model its data. Having an aerial LADAR image as data source, our aim is to extract a parametric description of the ground of our scenario in order to discern between the data samples that belong to the ground and those that belong to vehicles, objects or clutter. Once the samples are divided, we process each of the objects to perform an early classification refering to the object type (vehicle, building or clutter). The final step of our method is to estimate the pose of the interesting objects by building its corresponding oriented 3D bounding box.
Our method uses robust statistics in order to extract proper descriptions of both the ground and the oriented bounding boxes of the objects. Specifically, we use two robust parameter estimators : The Least Median Squares and the Variable Bandwith Quick Maximum Density Power Estimator, depending on the percentage of outliers that may be present in the different steps of our approach. Our method is open and can also be used along with other approaches that focus on extracting 3D invariant features or enhanced by applying a recognition step with the aid of model databases and 3D registration algorithms, such as the ICP.
Based on the requirements in several applications of object and surface remote identification, and considering the advantages of using multispectral techniques, several systems that allow image acquisition in both specific subbands and single wavelengths have been developed by our group. These systems are based on different techniques. They comprise visible and NIR ranges, with different spectral resolution. Three experimental setups have been developed. The first system is a camera with a filter wheel to choose different spectral bands. The second setup consists of a high-speed camera in which a 1 nm-resolution liquid crystal tunable filter has been assembled. The full system is automatic and allows a fast scan of visible subbands. The third setup uses the same imaging sensor as system #2, but in this case the filter has been substituted by a slit-spectrograph which splits the visible radiation into the different wavelengths that compose the small area observed. The desired wavelength is therefore selected by extracting the appropriate columns of the image acquired from the sensor. The correlation between wavelengths and the CCD array is determined in previous calibration steps. An additional rotatory stage allows the scanning of scenes. Software has been developed to control the systems and make automatic measurements. A new file format specially developed for this project allows the storage of all the images acquired in a single file, which allows a faster ulterior spectral analysis. A bands selection application simplifies the image acquisition depending on the observed scene. The images obtained by the systems will be analyzed in some subsequent stages: qualitative and behavioural study of the elements in the scene, comparison of resolution and operation capabilities of the different configurations and image calibration.
Active imaging systems allow obtaining data in more than two dimensions. In addition to the spatial information, these systems are able to provide the intensity distribution of one scene. From this data channel a certain number of physic magnitudes that show some features of the illuminated surface can be recovered. The different behaviours of the scene elements about the directionality of the optical radiation, wavelength or polarization improve the ability to discriminate them. In this work, the capabilities of one 3D imaging laser scanner have been tested from both dimensional and radiometric points of view. To do this, a simple model of the observing system and the scene, in which only the directional propagation of the energy is taken into account, has been developed. Selected parameters corresponding to transmission, reception and optomechanical components of the active imaging system describe the full sensor. The surfaces of a non-complex scene have been divided into different elements with a defined geometry and directional reflectance. In order to measure the directional reflectance of several materials in the specific wavelength where the laser scanner works, a laboratory bench has been developed. The calculation of the received signal by the sensor has been carried out using several radiative transfer models. These models were validated by experiments in a laboratory with controlled conditions of illumination and reflectance. To do this, a certain number of images (angle, angle, range and intensity) were acquired by a commercial laser scanner using several standard targets calibrated in geometry and directional reflectance.
This paper addresses the problem of tracking a target in an IR video sequence using a kernel based histogram representation of the target. In this field, gradient ascent methods have demonstrated useful results with weighted kernels and in particular Mean Shift is currently the most commonly used gradient scale method. Our approximation follows the work made by Hager, that uses a SSD objective function (derived from Matusita metric) and combines it with a Newton-like maximization method, resulting a fast gradient scale system. An important property is that this method enables the use of multiple kernels, allowing a more powerful representation with a minimum increasing of computational cost. We analyse the limitation of this representation using the Newton maximization algorithm and we introduce the concept of direction of ambiguity. This concept allows a criterion for choosing the kernels that drive the iteration to minimize the error criterion. The results we present show the improvements of the method over a tracking problem. The target is a small car with a great background similarity.