Search

To address the problem of extended target tracking (ETT) with measurement-origin uncertainty in a multipath environment with terrain-constrained motion model, a new generalized version of the standard probabilistic data association (PDA) filter, called MP-ET-PDA, based on random matrices (RM) is proposed in this paper. In the MP-ET-PDA filter, we assume that multipath detections and clutter are possible in the extended target tracking problem, which are prevalent in practical systems but barely addressed in the literature. Further, a clustering-aided MP-ET-PDA algorithm with a reduced computational complexity that makes use of the Variational Bayesian (VB) technique, called MP-ET-PDA-VB, is presented to provide near real-time processing capability in ETT problems with an uncertain multipath environment. In addition to using a constant velocity motion model, a new terrain-constrained motion model is presented for scenarios where terrain-following is required by motion-constrained autonomous vehicles. The posterior Cramer-Rao lower bound (PCRLB), which quantifies the best possible accuracy in realistic ETT problems with multipath detections and measurement-origin uncertainty, is derived as the benchmark for performance evaluation. Simulations results demonstrate the improved performance of the proposed algorithms.

Accurate georegistration is required in airborne surveillance of ground targets since even a small bias in sensor registration may result in a large error in the converted target geolocation. Thus, this article addresses the problem of estimating and correcting the sensor biases in target geolocation. The effect of sensor biases on target geolocation is modeled and decoupled from the true target geolocation. By assuming that the biases are small, the effect of biases is separately addressed by a linear term in the converted target geolocation. Based on this model, two computationally efficient methods are proposed to estimate the biases. The converted target geolocation can then be corrected based on the bias estimates. Simulation results demonstrate the effectiveness of the proposed methods in estimating sensor biases and correcting target geolocations as well as their high computational efficiency.

Aerial images and videos are extensively used for object detection and target tracking. However, due to the presence of thin clouds, haze or smoke from buildings, the processing of aerial data can be challenging. Existing single-image dehazing methods that work on ground-to-ground images, do not perform well on aerial images. Moreover, current dehazing methods are not capable for real-time processing. In this paper, a new end-to-end aerial image dehazing method using a deep convolutional autoencoder is proposed. Using the convolutional autoencoder, the dehazing problem is divided into two parts, namely, encoder, which aims extract important features to dehaze hazy regions and decoder, which aims to reconstruct the dehazed image using the down-sampled image received from the encoder. In this proposed method, we also exploit the superpixels in two different scales to generate synthetic thin cloud data to train our network. Since this network is trained in an end-to-end manner, in the test phase, for each input hazy aerial image, the proposed algorithm outputs a dehazed version without requiring any other information such as transmission map or atmospheric light value. With the proposed method, hazy regions are dehazed and objects within hazy regions become more visible while the contrast of non-hazy regions is increased. Experimental results on synthetic and real hazy aerial images demonstrate the superiority of the proposed method compared to existing dehazing methods in terms of quality and speed.

Geo-registration and geo-location of data collected by video sensors such as electro-optical and infrared cameras are two fundamental steps in the airborne surveillance of ground targets. With the availability of high-resolution imaging sensors and detailed mapping or terrain data sources, video data plays an increasingly important role in modern surveillance platforms like unmanned aerial vehicles and airborne, ground, or maritime surveillance systems. Surveillance systems without any compensation for the inevitable sensor registration errors, i.e., biases, may make geo-location erroneous and render the surveillance platform less effective for precision targeting. This article deals with the modeling of sensor biases in geo-location and proposes a method to estimate them. The proposed method leads to a minimization problem with a nonlinear cost function. Detailed derivation of the bias model is given along with an algorithm to find the bias parameters. The achievable lower bounds for debiased geo-location are provided and simulations are used to demonstrate the validity of the proposed method.

Temporal registration of sensors is an essential preprocessing step in multisensor target tracking systems. A new approach for multisensor time-offset estimation is proposed in this article. First, the time offset pseudomeasurement equation is derived and calculated in both centralized and decentralized scenarios, where measurements and local tracks are available at the fusion center, respectively. The observability of time offset is analyzed theoretically, which shows that only relative time offsets between sensors are observable. Second, a two-sensor two-stage filtering method is developed with four different formulations corresponding to different time-offset statistical models and target dynamic models to obtain a relative time-offset estimate. A multisensor two-stage filter is also proposed to obtain a minimum-bias time-offset estimate. Furthermore, the interacting multiple model estimator is used to deal with temporal registration in the presence of target maneuvers. Finally, the posterior Cramer–Rao lower bound (PCRLB) is derived for relative time-offset estimation. Simulation results show that the proposed algorithm with two sensors yields an empirically unbiased estimate of the relative time offset, and that the root-mean-square errors (RMSEs) match the corresponding PCRLB. Simulation results for multisensor target tracking are also presented to demonstrate the validity of the proposed algorithms.