Adaptive visual tracking based on Discriminative Feature Selection for Mobile Robot

The main challenges of visual tracking for mobile robot come from variation of target's appearance and disturbance of environment, such as pose changes of target, illumination changes, and cluttered background. This paper presents a robust adaptive visual tracker which is able to capture the varying appearance of target under different environments without gradual drift. We propose a novel and flexible feature space evaluation function which is formed by the weighted sum of two components: the similarity measure and the discriminating ability measure. To minimize the influence of background, a new salient feature selection mechanism is proposed to clearly distinguish between target and background. A novel target model updating mechanism is introduced to avoid gradual model drift with time, and a pure, adaptive and time-continuous target model is obtained for each input frame without off-line training and prior knowledge. The proposed discriminative feature selection and target model updating mechanism is embedded in a Mean-shift tracking system which iteratively finds the nearest local optimal localization of target. Experimental results on a mobile robot system demonstrate the robust performance of the proposed algorithm under different challenging conditions.