Video object retrieval mechanism with the effective video tracking model based on Deep long-short term memory

Abstract

In the field of modern intelligent systems, the video objects retrieval is still challenging due to object confusion, different posing, same appearance between the objects, small object size, and the interactions among the multiple objects. To tackle the above challenges, this paper presents the novel approach for increasing the performance of the retrieved video object using the developed Deep long-short term memory (Deep LSTM). Initially, the key frames from the video is acquired and the objects is detected from the key frames on the basis of Nearest Neighbourhood Algorithm (NSA). Then, the trajectory of detected objects is tracked using Deep LSTM such that the location is tracked in the video. Once the video objects are tracked, the object retrieval mechanism is enabled in the model that is done using Holoentropy-based weighed Query specific distance (holoentropy-based weighted QSD) that is modelled with the help of holoentropy measure. Thus, the video object is tracked and retrieved effectively from video. The effectiveness of proposed Deep LSTM is computed which revealed maximal precision of 0.979, recall of 0.919, F-measure of 0.914 and MOTP of 0.933.