Recognition of human and animal movement using infrared video streams

Distinguishing human motion from animal motion is important in many applications using infrared video streams, such as surveillance systems for homeland security and collision avoidance systems for nighttime driving safety. In this paper we present a technique to distinguish human motion from animal motion using infrared video sequences. In our technique, we uses frame differencing to represent object motion. Space-time correlation is used to characterize different type of motions. Our motion features are defined by Renyi entropy and mean values calculated from the correlations. A support vector machine-based classifier is used to classify the motion features. Our experimental results show that our technique is quite effective at distinguishing human motion from animal motion using infrared video sequences.