JD AI and Beijing University researchers have recently developed PVSS, a progressive vehicle search system for surveillance video networks. This system helps in effectively searching a particular vehicle which appears in surveillance footage.
Generally, vehicle search systems comprise of various useful applications such as automated surveillance and smarter transportation. Consequently, such systems allow users in searching time interval and area for finding out vehicle’s location at different times. Furthermore, this existing methods require cropped images from the videos to identify the license plate number of the targeted vehicle. Moreover, this vehicle search approaches known as vehicle re-identification as it mainly target on content-based vehicle matching.
However, this vehicle searching method is challenging in the area of inter-instance differences between similar vehicles and in different cameras. In addition to that, due to low resolution and noise, users often misrecognizes license plates in surveillance images.
PVSS Showed State-of-the-art Results On Large Scale Vehicle Search data set
Researchers’ new vehicle search system covers up all the problems such as representation, vehicle detection, indexing, matching, and storage. This new system composed of three key modules such as crawler vehicle data, multi-grained vehicle indexer, and progressive vehicle searchers.
A series of data structures in this new vehicle search system ensures high accuracy and efficiency during search. The crawler not only helps in extracting visual contents but also contextual information from surveillance networks. A deep learning-based model exploits the multimodal data to obtain robust features of vehicles. Subsequently, the vehicle indexer extracts multi-grained attributes of the vehicles such as license plate fingerprints.
This progressive vehicle search method outperforms in all the multi-modal methods and appearance-only search methods.
