Research On Key Technology Of Infrared Target Detection In Large Field Of View System

In modern high-tech warfare, infrared imaging detection has become the important complementary or alternative of radar for target surveillance and early warning, which is also the main way of battlefield defense under asymmetric warfare environment in future. At present, the main direction of developing third generation of infrared detection system is to use large size of focal plane array(FPA) with thousand pixels, which implements target detection and recognition by long-range, large field of view(FOV) and high-resolution. These systems have the capabilities of high-precision detection, fast scanning, horizontal 360-degree panoramic coverage and large pitch angle searching. Thus, problems such as infrared image processing of large FOV, real-time target detection are highlighted. Based on the support of equipment pre-research plans, and application requirement of new generation high-performance infrared search system, this paper addresses the target detection technology in large FOV infrared image.There are four parts in this thesis:For the infrared imaging systems with high sampling width applying to the traditional display device or real-time processing system with 8bits data width, the paper presents a new high dynamic range compression and detail enhancement(DRCDDE) algorithm for infrared image. First, a bilateral filter is adoppted to separate the original image into two parts: the base component that contains large-scale signal variations, and the detail component that contains high-frequency information. Then, the operator model for DRC with local contrast preservation is established, along with a new proposed nonlinear intensity transfer function(ITF) to implement adaptive DRC of base component. For detail component, depending on the local statistical characteristic, we set up suitable intensity levels extension criterion to enhance the low-contrast details and suppress noise. Finally, results of the two components are recombined with a weighted coefficient. Experiment results by real infrared data, and quantitative comparison with other well-established methods show the better performance of the proposed algorithm. Furthermore, the technique could effectively project dim target while suppress noise, which is beneficial to image display and target detection.In the panoramic field-of-view(FOV) infrared imaging search system(PIRSS), the infrared image background was especially complicated, along with rapidly increasing of the data quantity. According to the infrared image’s characteristics in the PIRSS, a flow of detecting algorithm based on the weighted local entropy(WLE) matrix of image blocks was proposed in this paper. Firstly, it established the image blocks matrix for the entire image, which was based on the spatial distributing characteristics of the panoramic image. Then the paper presented a new characteristic function called weighted local entropy, and calculated the WLE matrix for the image blocks. Finally, an appropriate adaptive threshold method based on the analysis of WLE matrix was adopted, which implemented the region separation of candidate targets from background and obtained the ROI. Experimental results demonstrated that the proposed algorithm was effective and befitting for the infrared target detection in large FOV. It also has good performance for real-time processing and engineering realization.Infrared target detection is one of the most important techniques in computer vision, particularly of great significance in the infrared search and track for self-defense. This paper proposes an algorithm flow for typical infrared target positioning and automatic scale selection with multi-scale analysis. In addition to as the pre-detection stage, it also provides scale information for the target without any prior knowledge. Firstly, the plane target is modeled by blob-like structure feature. Based on Lindeberg’s axioms, we numerically implement the construction of two-dimensional discrete scale space for infrared images. Then, through introducing the concept of discrete derivative approximations, a robust differential entity is derived to extract the blob-like feature in discrete scale space(DSS), named discrete analogue of the Laplacian of the Gaussian(NDALo G). Detecting the NDALo G operator’s maxima of its 26-point neighborhood within the DSS, keypoints of the image are achieved initially. Eventually, the criteria for redundant keypoints inhibition are established by analyzing multi-scale singular singularity of target, so as to implement extraction and scale selection of the interest points. The experiments on simulated data and real infrared image with different types of backgrounds and target scales validate the detection and scale selection capabilities of the proposed algorithm. Experimental evaluation results show that our method outperforms the state-of-the-art methods in terms of detection accuracy and low false-alarm rate.Due to the real-time processing and high-speed data transfer, we propose a high-performance processing platform based on field programmable gate array(FPGA) + multicore digital signal processor(DSP) architecture. To reduce the platform’s power consumption, dynamic power monitoring technology is used to provide variable core voltage management for DSP. Then mapping an application to the platform is illustrated, which is guided by parallel-task segmentation mechanism of multi cores system. We show that, the designed platform with the optimization strategies achieves powerful capabilities of data throughput and processing performance.