Online learning neural tracker

Abstract

Object tracking is a fundamental computer vision problem and is required for many high-level tasks such as activity recognition, behavior analysis and surveillance. The main challenge in the object tracking problem is the dynamic change in object/background appearance, illumination, shape and occlusion. We present an online learning neural tracker (OLNT) to differentiate the object from the background and also adapt to changes in object/background dynamics. In the proposed tracking system, we propose a new mobile object detection module which identifies new mobile objects in the scene and then OLNT faithfully locates them in subsequent frames. The OLNT extracts region-based features like region-based color moments for larger mobile objects and color and texture features at pixel level for smaller mobile objects.For target modeling and object tracking, new neural classifier algorithm based on risk sensitive loss function is proposed to handle issues related to sample imbalance and change in characteristic of object class in future frames. The proposed neural classifier automatically determines the number of neurons required to estimate the posterior probability map. In the online learning neural classifier, only one neuron parameter is updated per tracker to reduce the computational burden during online adaptation.The tracked object is represented using an estimated posterior probability map. The signature of the posterior probability map is used to adapt the bounding box to handle the scale change and improper initialization.For illustrating the advantage of the proposed OLNT under rapid illumination variation, change in appearance, scale/size change, and occlusion, we present results from benchmark video sequences. The results clearly highlight the advantages of the proposed OLNT. Finally, we also present the comparison with well-known ensemble tracker in one the sequence and highlight the advantage of the proposed tracker. Index TermsObject tracking, neural classifier, Gaussian activation function, posterior probability map, signature in 2D, risk sensitive hinge loss function.

Object Tracking Based on Compressive Features and Extreme Learning Machine

“…This method uses information from the first frame to train the NN with no further model updating. Suresh et al presented an online learning neural tracker with radial basis function network, and the learning algorithm updates the parameters of the nearest neuron to reduce the computational complexity [30]. Serratosa et al exploited a probabilistic integrated object recognition and tracking framework with respectively Bayesian method and a multilayer perceptron, and their test results showed that the perceptron method outperforms the Bayesian [31].…”

Planning and relaxed state EEG signal classification using complex valued neural classifier for brain computer interface

“…Nowadays, neural networks are applied in various domains such as classification problems [1]- [9] and for prediction [10] [11], Control system [12]- [14], workstations [15], aeronautical engineering [16] and human-machine interface [17]. Recently, multidisciplinary Brain Computer Interface researches are taking place in a rapid pace for providing communication abilities for people with rigorous motor disabilities.…”

A Meta-cognitive Fully Complex valued Fast Learning Predictor network for solving real valued prediction problems

“…The domain of artificial neural network are applied for classification problems [1] and for prediction, Control systems [2][3]etc.,. Recently, the data in the applications like signal processing [4], [5], image processing [6] and communication [7], etc., are naturally complex-valued in nature and varies with time.…”