A Novel Approach for Real-Time Vehicle Re-identification Using Content-Based Image Retrieval with Relevance Feedback

Abstract

Automated smart traffic surveillance systems constitute a significant part of smart city environments and have attracted significant research attention in recent years. Vehicle re-identification is a major challenge in automated traffic surveillance systems in smart city environments. Vehicle re-identification is the process of retrieving instances of the target vehicle given a gallery of numerous vehicle images. Though multiple models were proposed to perform the task of vehicle re-identification, the models struggle in terms of real-world implementation because of their complexity and computational requirements. This is mainly due to the focus on computation-heavy feature extraction processes, along with complex pre-processing and post-processing steps. To address these issues, an approach incorporating content-based image retrieval techniques with deep neural models that are computationally efficient is proposed. The approach also considers relevance feedback during the post-processing phase. Experimental results reveal that the incorporation of relevance feedback technique as a post-processing technique in vehicle re-identification helps achieve significant improvement in terms of mean average precision and Rank@k.