Object Tracking Based On Biologically Inspired Model

Visual tracking is an important issue in the field of computer vision, which has broader prospects for human Computer interaction, video surveillance, automatic driving. Although great progress has been made in recent years, it is still difficult to track a general object in wild environments, due to influences by object deformation, scale change, illumination change and occlusion. Researchers are trying to solve these problems from appearance model and tracking model.A number of object tracker have been proposed, but most of them have their strengths and weaknesses, and could not handle all challenging situations. Given the superior tracking performance of human visual system(HVS), an ideal design of biologically inspired model is expected to improve computer visual tracking. This is however a difficult task due to the incomplete understanding of neurons' working mechanism in HVS. This paper aims to address this challenge based on the analysis of visual cognitive mechanism of the ventral stream in the visual cortex, which simulates shallow neurons(S1 units and C1 units) to extract low-level biologically inspired features for the target appearance and imitates an advanced learning mechanism(S2 units and C2 units) to combine generative and discriminative models for target location. In addition, fast Gabor approximation(FGA) and fast Fourier transform(FFT) are adopted for realtime learning and detection in this framework. Extensive experiments on large-scale benchmark datasets show that the proposed biologically inspired tracker(BIT) performs favorably(81.7%)against state-of-the-art methods on TB50. The acceleration technique in particular ensures that BIT maintains a speed of approximately 45 frames per second.In addition, there are two aspects to improve bio-inspired appearance model(BIAM) and bio-inspired tracking model(BITM). First, a deep convolution feature is used to replace S1 and C1 units, and improves the tracking robustness. Second, a scale estimation method is only used to estimate bounding box size independently. On the other hand, a fast scale feature is designed based on Spatial Pyramid Pooling(SPP). Comprehensive experiments show that, two methods above can effectively improve the performance of BIT. The BIT based deep learning achieves 84.9% accuracy on TB50; the mean F-score of BIT based on scale estimation is 0.724 on ALOV300++.This thesis exploits a novel research field for object tracking. BIT would be helpful for the research of appearance model and tracking model, and promote the employment and exploration of visual perception theory in computer vision.