Person Re-identification Using Human Gait Signature

Abstract

An automated surveillance environment is constituted using personal re-identification as an integral part. Usually, this process is carried out using vision sensors by checking for the physical appearance of the person that is their color etc. thus putting constraints in the process of re-identification. This visual based identification comprises a lot of constraints suggesting the researchers to go for further research finding other ways and they get to the point of using human gait as a replacement which is much more efficient and reliable than using visual based analysis in which results may get affected by the surroundings. Already research has been done in this field, in this research extended work and is now proposed by using LSTM model over three sensor modalities. Exper iments are conducted over multiple recurrent layers, activation functions and dropout values for the step data. From these experiments we analyzed that using the less re current layers for the step data model gives almost the same accuracy as the previous study but, in less time, which reduces the time complexity of the model thus results in an efficient model.