Pedestrian detection is the major task of many infrared surveillance system. Due to the technical limitation of sensor or the high cost of advanced hardware, the resolution of infrared images is usually low, which is not capable of meeting the high quality requirement of various applications. Compressed sensing capturing and represents compressible signals at a sample rate significantly below the Nyquist rate, is considered as a new framework for signal reconstruction based on the sparsity and compressibility. Thus, the compressed sensing theory enlightens a computational way to reconstruct a high resolution image on the basis of a sparse signal, i.e. the low resolution image. The proposed method use low resolution and high resolution infrared pedestrian images to train an over-complete dictionary through K-SVD algorithm, by which the pedestrian are sparsely well-represented. Two distant infrared cameras in the same scene are used to capture high and low resolution image to make sure same pedestrian pair is sparsely represented under the over-complete dictionary. Therefore the similarities are learning between input low resolution image patches and high resolution image patches. The popular greedy algorithm Orthogonal Matching Pursuit (OMP) is utilized for sparse reconstruction, providing optimal performance and guaranteeing less computational cost and storage. We evaluate the quality of reconstructed image employing root mean square error and peak signal to noise. The experimental results show that the reconstructed images preserve wealthy detailed information of pedestrian, and have low RMSE and high PSNR, which are superior to the traditional super-resolution methodologies.
This paper proposed a fast human action recognition algorithm which utilized two features that can be described as iconic posture and fast moving. At first, a human detection algorithm is used to detect human object in every frame. Then regions marked as human are sent into a trained deep classification network to match trained iconic postures in key frame. Then several frames before key frame and after key frame are examined by frame differences, which are used to compensate background movement and perform human motion speed judgment. After the key frame pinning and speed judgment, the final recognition results are determined.
Hyperspectral imaging sensors can acquire images in hundreds of continuous narrow spectral bands. Therefore each object presented in the image can be identified from their spectral response. However, such kind of imaging brings a huge amount of data, which requires transmission, processing, and storage resources for both airborne and space borne imaging. Due to the high volume of hyperspectral image data, the exploration of compression strategies has received a lot of attention in recent years. Compression of hyperspectral data cubes is an effective solution for these problems. Lossless compression of the hyperspectral data usually results in low compression ratio, which may not meet the available resources; on the other hand, lossy compression may give the desired ratio, but with a significant degradation effect on object identification performance of the hyperspectral data. Moreover, most hyperspectral data compression techniques exploits the similarities in spectral dimensions; which requires bands reordering or regrouping, to make use of the spectral redundancy. In this paper, we explored the spectral cross correlation between different bands, and proposed an adaptive band selection method to obtain the spectral bands which contain most of the information of the acquired hyperspectral data cube. The proposed method mainly consist three steps: First, the algorithm decomposes the original hyperspectral imagery into a series of subspaces based on the hyper correlation matrix of the hyperspectral images between different bands. And then the Wavelet-based algorithm is applied to the each subspaces. At last the PCA method is applied to the wavelet coefficients to produce the chosen number of components. The performance of the proposed method was tested by using ISODATA classification method.
Small target is also weak target, which is likely to be a threat to the observation platform. So small target detection is an important task for many automatic object detection system. Otherwise, small target detection is a challenge for many complex scenes because of the low SNR and sophisticated background. This paper introduced a fast and effective method for small target detection in infrared scene with complex background, which is suitable for missile guidance and menace warning. Firstly, a template is created to detect the local maxima in the image. Secondly, a constrained double criteria region growth algorithm is performed to form separate regions. Finally, extracted regions are selected by a small round target filter, after which, the remaining connected regions are considered to be detected small targets. The proposed algorithm was applied on videos captured by cooled infrared imagers. Experimental results show the method introduced in this paper is efficient and effective, which is suitable for time sensitive automatic target detection.
The spectral characteristics of infrared radiation from target provide significant characteristics information for target's detection and track including radiance brightness, radiance intensity and spectrum characteristics of target. And the same time, the spectral characteristics provide the basis of target detection and recognize equipment's waveband optimization design and detection capability analysis. This paper using the passive imaging Fourier transformation infrared spectrometer measure the infrared spectral characteristic of target. The spectral range cover the medium wave and long wave infrared. And the instrument can interference imaging in 320×256 spatial resolution or other window size. This paper designs a set of calibration and test processes to realize the infrared spectral radiance measurement of target. Using this method, this paper test some typical infrared target. After the radiance calibration, the calibrated result is verified by standard radiance source. Thereby, the remote measurement of infrared background is taken as the comparison test. Finally, the typical infrared target spectral features are extracted and measured. The test results show that the method mentioned in this paper is practical.
In infrared image, sea-level line could be hard to distinguish because of noises caused by wave clutters and sunlight conditions.This paper proposed a fast sea-level line extraction method which could localize the sea-level line in complex infrared sea-sky scenes. First, a down sample operation was performed to obtain a low resolution image which could reduce the time consumption without blurring the sea-level line, and then the Canny edge detection was carried out to extract edges in the scene. Second, the intersecting edges were separated by removing the joints of edges according to a certain rule, and the bounding rectangle of every short edge was obtained which helped to select straight lines, and then a long edge segmentation operation was used to count in possible sea-level line. Third, a line concatenation method was performed by their slopes and intercepts comparison. Finally, for sea-level line verification, the second-order vertical grads are calculated in the two sides of possible sea-level line. Experiments show that the proposed method is fast and effective for various kinds of infrared sea-sky scenes, and it is feasible even for the scenes where the sea-level line is blurring and hard to distinguish.
Usually, there is a distinguishable sea-level line in the infrared sea image, where many possible objects can be found.
While relative to varies kinds of objects, the sea-level line can be more easily detected, which makes the sea-level line
detection a important step in object detection and recognition in infrared sea images. This paper proposed a fast sea-level
line detection method, which estimated pixels of sea-level line through the gray characteristic of neighborhood of them,
performed a preliminary sea-level line positioning by line fitting, and verified the results by the linear feature of sea
surface edges. Based on the results of sea-level line detection, a fast object candidate detection method was introduced.
Experimental results proved that the existing and position of sea-level line can be determined and preliminary object
detection can be performed by the proposed method.
Since the infrared imaging system has played a significant role in the military self-defense system and fire control system, the radiation signature of IR target becomes an important topic in IR imaging application technology. IR target signature can be applied in target identification, especially for small and dim targets, as well as the target IR thermal design. To research and analyze the targets IR signature systematically, a practical and experimental project is processed under different backgrounds and conditions. An infrared radiation acquisition system based on a MWIR cooled thermal imager and a LWIR cooled thermal imager is developed to capture the digital infrared images. Furthermore, some instruments are introduced to provide other parameters. According to the original image data and the related parameters in a certain scene, the IR signature of interested target scene can be calculated. Different background and targets are measured with this approach, and a comparison experiment analysis shall be presented in this paper as an example. This practical experiment has proved the validation of this research work, and it is useful in detection performance evaluation and further target identification research.
The existence of non-uniformities in the responsitivity of the element array is a severe problem typical to common infrared detector. These non-uniformities result in a “curtain’’ like fixed pattern noises (FPN) that appear in the image. Some random noise can be restrained by the method kind of equalization method. But the fixed pattern noise can only be removed by .non uniformity correction method. The produce of non uniformities of detector array is the combined action of infrared detector array, readout circuit, semiconductor device performance, the amplifier circuit and optical system. Conventional linear correction techniques require costly recalibration due to the drift of the detector or changes in temperature. Therefore, an adaptive non-uniformity method is needed to solve this problem. A lot factors including detectors and environment conditions variety are considered to analyze and conduct the cause of detector drift. Several experiments are designed to verify the guess. Based on the experiments, an adaptive non-uniformity correction method is put forward in this paper. The strength of this method lies in its simplicity and low computational complexity. Extensive experimental results demonstrate the disadvantage of traditional non-uniformity correct method is conquered by the proposed scheme.
Since infrared image quality depends on many factors such as optical performance and electrical noise of thermal imager, image quality evaluation becomes an important issue which can conduce to both image processing afterward and capability improving of thermal imager. There are two ways of infrared image quality evaluation, with or without reference image. For real-time thermal image, the method without reference image is preferred because it is difficult to get a standard image. Although there are various kinds of methods for evaluation, there is no general metric for image quality evaluation. This paper introduces a novel method to evaluate infrared image without reference image from five aspects: noise, clarity, information volume and levels, information in frequency domain and the capability of automatic target recognition. Generally, the basic image quality is obtained from the first four aspects, and the quality of target is acquired from the last aspect. The proposed method is tested on several infrared images captured by different thermal imagers. Calculate the indicators and compare with human vision results. The evaluation shows that this method successfully describes the characteristics of infrared image and the result is consistent with human vision system.
This paper proposed a smoke detection method suitable for constant speed rotating platform. The movements of long
range scenes in images acquired by camera on rotating surveillance platform is approximate translation, the proposed
method uses a grid point based image patch matching to obtain the parameters of the translation, and compensates the
movements of scenes to generate short-time stable scenes for smoke detection. After the movement compensation of the
video, the smoke candidate regions are selected by two-stage background difference, color judgment and shadow
judgment. The two-stage background difference which takes advantage of fast background updating and slow
background updating is performed to detect slowly changing regions. The color judgment is applied to filter out
non-smoke color regions, for smoke always takes on gray colors. And the shadow judgment is to carry out by the light
color feature and the no texture changing feature of shadows. An image buffer pool is established for smoke diffusion
analysis where forward region trajectories and corresponding backward region trajectories are obtained. Those regions’
moving trajectories are obtained by a consecutive frame region matching method based on their corresponding position
distance and are filtered by a consistency constraint, after which, the translation, expanding, shrinking features of
remaining smoke candidate regions are attained, which are used for movements and diffusion judgments. The smoke
candidate regions passing all of above judgments are considered smoke regions. Experimental results show the proposed
method is prominent for forest smoke detection.
With constraints to the performance of the IR detector, IR image usually has lower visual effect with low contrast and less detailed information. In this paper, a new dynamic range infrared image details enhancement algorithm is studied, using a bilateral filter to extract a base component and a detail component. Then these two components are compressed to fit the display dynamic range and then recombined to obtain the output-enhancement image. This algorithm has solved the problem of ripple phenomenon which exists in the traditional infrared image digital detail enhancement. Finally, the algorithm described in this paper is proved experimentally that can provide better DDE effect.
Fire detection based on video surveillance is a very effective method for large area outdoor fire prevention, but the
unpredictable place and time makes automatic fire detection a difficult problem. This paper adopts a loose color
selection and frame differential to narrow down possible fire regions, where every pixel’s temporal color variations
are analyzed by 3-state Markov modals. One of the Markov modal is used for brightness variation examination and
the other one is used for fire color likeness that is measured by color difference. In order to eliminate false
detections, the fractal dimension calculation and texture match are performed. Experimental results prove the
proposed method is feasible and suitable for outdoor or indoor fire detection in surveillance videos.
This paper introduces a novel vehicle detection method combined with probability voting based hypothesis generation
(HG) and SVM based hypothesis verification (HV) specialized for the complex background airborne traffic video. In HG
stage, a statistic based road area extraction method is applied and the lane marks are eliminated. Remained areas are
clustered, and then the canny algorithm is performed to detect edges in clustered areas. A voting strategy is designed to
detect rectangle objects in the scene. In HV stage, every possible vehicle area is rotated to align the vehicle along the
vertical direction, and the vertical and horizontal gradients of them are calculated. SVM is adopted to classify vehicle
and non-vehicle. The proposed method has been applied to several traffic scenes, and the experiment results show it’s
effective and veracious for the vehicle detection.