Target tracking with particle filter based on multiple cues fusion

To overcome the shortcoming of single visual cue in complex environments, a tracking algorithm based on the multi-cue adaptive fusion is proposed. The color cue and the texture based on the discrete wavelet transform (DWT) are utilized to describe the target. When fusing each cue, a fusion scheme based on democratic integration is applied. The fusion scheme adaptively adjusts the weight of each cue according to the current tracking situations, which increases the reliability of observation and improves the robustness of observation model. Due to the use of reliable cues for tracking, the failure of single cue in complex scenes is avoided. Meanwhile, when designing the likelihood function of observation, the noise variance of each cue is updating to increase the discrimination of each cue. During designing particle filter based tracking algorithm, the likelihood model is constructed depending on adaptive cue fusion mechanism, thus enhancing the robustness of tracking algorithm. Through experiments, the result of better tracking are obtained in circumstances that the target is arbitrarily moving or rotating, partially or completely occlusion, as well as the light changes. When the target with similar color appears, the tracking algorithm with the texture model can distinguish the target, thus addressing the collision problem of similar targets.