Software Design And Algorithm Research Of Infrared Target Detection Based On Multi-core DSP

Infrared target detection that plays an essential part in precision guidance,air defense and early warning,assisted driving,and intelligent surveillance is one of the research hotspots in computer vision.The accuracy and efficiency of target detection are difficult to be balanced by the computing power and memory resources of embedded platforms.This thesis explores the infrared target detection algorithm and software design based on the multi-core high-performance TMS320C6678 DSP about the problem.By using multi-core parallelism and instruction set optimization,the accuracy and efficiency of infrared target detection are effectively improved under resource-constrained conditions.Firstly,starting from the task of infrared target detection under resource-constrained conditions,this theis analyzes the characteristics of infrared target imaging,clarifies the difficulties of infrared target detection algorithms,and makes a requirement analysis and overall design for the algorithm part.Then the software design is analyzed in conjunction with the overall hardware architecture of the system,and the functions to be accomplished by the TMS320C6678 are clarified.Next,the algorithm part is explored in this thesis.At first,the images are preprocessed according to different noise types.Secondly,the classical target detection algorithm based on the dominant orientation templates is analyzed,whose the advantage of the algorithm is that it can build the training set with a small number of templates and use binary encoding to represent the features,which greatly accelerates the process of template matching.Thirdly the algorithm is improved for the limitations of the algorithm under infrared target detection in two aspects,one is to solve the problem of low detection accuracy of the original algorithm when the gradient features of the infrared image are not significant by incorporating the grayscale features into the main gradient features,which greatly improves the discriminative ability of the features.The second point solves the problem of poor robustness of the dominant orientation templates in dealing with partial occlusion by a method improving the process of dominant orientation templates matching based on connected-component labeling,which greatly improves the detection accuracy of the target when it encounters partial occlusion.Finally,software design and algorithm porting optimization are made in TMS320C6678 platform.The software design mainly includes DSP interrupt configuration,clock configuration,memory configuration,software configuration of each communication interface driver and DSP self-start implementation.The algorithm implementation part successfully transplants the target detection algorithm proposed in this thesis to the image processing board,and makes deep optimization including instruction set optimization,multi-core parallel optimization and other optimization methods based on TMS320C6678 platform.At last,after experimental testing,it can not only be obtained that the detection accuracy of the image processing board is above 90%,but the detection time of one frame is within 40 ms,which can accurately detect infrared targets in real time.