Target Reconstruction And Tracking Algorithm Research Dased On Compressive Sensing Theory

In the field of target tracking, deformation, shade, lighting, scale, rotation, clutter background is several factors affecting the effects of trace.For non-rigid object, in the process of movement, the occurrence of deformation is often happened. Due to the fixed camera position, object is easy to be kept out by other barriers, lead to the loss of tracking.From indoor to outdoor, illumination change is very severe, this can lead to radical changes of tracking target characteristics.When the tracked objects are away from or close to the camera, the object will be scaled, and this can also lead to the loss of the tracked object.When objects rotate, the characteristics of traking object will change a lot.If background is complex, the tracking target and the background are very hard to distinguish,this will influence the effect of the track.Early tracking algorithm, based on the feature extraction of tracking, in the absence of complex changes, can achieve good results.In the practical application, daily living environment differs in thousands ways, all kinds of interference may occur, and that will seriously affect the effect of the track. So the tracking algorithm is not simple tracking, but the combination of tracking and detection. In recent years, some excellent tracking algorithms are using the machine learning algorithms, to study the characteristic of classification, in order to achieve good adaptability, in this brief, for tracking objects, through the use of compressive sensing algorithm, dictionary is established beforehand, the tarcking target is denoised and so that can adapt to all kinds of possible interference target, effectively improve the tracking accuracy.In this thesis, because of the shortcoming of particle filter for large time-consuming on features extraction, a histogram features fast computing hardware architecture is proposed.The hardware architecture can be implemented on FPGA, the GPU, and it can greatly save the integral histogram initialization time, accelerate the calculation of histogram feature. In addition, this paper on the basis of the follow-up study, the algorithm can be applied to specific time scenario, we design a fall alarm and anomaly detection system based on raspberry pie, providing guarantee for the elderly live alone.