An Improvement of Kernel-Based Object Tracking Based on Human Perception

Abstract

The objective of the paper is to embed perception rules into the kernel-based target tracking algorithm and to evaluate to what extent these rules are able to improve the original tracking algorithm, without any additional computational cost. To this aim, the target is represented through features that are related to its visual appearance; then, it is tracked in subsequent frames using a metric that, again, correlates well with the human visual perception (HVP). The use of HVP rules are twofold advantageous: it allows us to both increase tracking efficacy and considerably reduce the computational cost of the tracking process-thanks to the reduced size of the perceptual feature space. Various tests on video sequences have shown the stability and the robustness of the proposed framework, also in the presence of both other moving objects and partial or complete target occlusion in a limited number of subsequent frames.

Classifying With Adaptive Hyper-Spheres: An Incremental Classifier Based on Competitive Learning

“…(1)For Each c i in C (2) For Each c j in C except c i (3) c big = max radius (c i , c j ), c small = min radius (c i , c j ),d t = d(c big , c small ); (4) If d t + r small <= r big + θ × r small //θ is the merging coefficient (5) Merge c i and c j ; (6) End If (7) End For (8)End For where d safe represents a safe distance at which a hypersphere should be from the closest DMZ point. The logics of this stage are outlined in Algorithm 1 below.…”

“…End If (6) Else (7) Mark c i as "core hyper-sphere"; (8) End If (9)End For 1) The first type of hyper-spheres includes large number of instances. Because these are the fundamental hyperspheres that contain most data points, they are marked as "core hyper-spheres."…”

Kernel‐Based Multiview Joint Sparse Coding for Image Annotation

“…Experimental results on three widely used datasets show the effectiveness of our proposed method compared to the related methods for image annotation tasks. Considering that some non-ground-truth annotations can still describe the image well in our experiments, for the future research, we try to further involve human perception [33] to measure the relevancy between the non-ground-truth annotations and the images to show the potential flexibility of our proposed framework. Besides, we are going to apply deep learning features [10,34] and multiple kernel learning [18] into our framework.…”

Automated sewer pipe defect tracking in CCTV videos based on defect detection and metric learning

“…MOT has been an important research topic in the field of computer vision due to its potential applications in various areas, such as video surveillance, autonomous cars and crowd behavior analysis [50]. There have been different frameworks developed for visual tracking, such as discriminative correlation filters (DCF) [51,52], silhouette tracking [53,54], Kernel tracking [55,56] as well as point tracking [57]. However, the performances of the aforementioned methods are not satisfactory as expected due to the handcrafted feature representations e.g.…”