Multiple Object Detection and Tracking with Scene Context

Image based trackers may fail in real world scenarios due to the performance limitations of object detectors that generate noisy observations under illumination changes, reflections and occlusions. For this a new framework is proposed with a feedback from scene contextual information. Thus it is possible to overcome all the illumination variations and can identify the missing detection. This achieved through a Probability Hypothesis Density (PHD) filter that spatially modulates its strength based on the learned contextual information. While applying a patch-based technique to this it is able to overcome both spatial and temporal consistency of the static background. Through this it is possible to reconstruct the background images by picking up the original colour.