Browse Prior Art Database

System and apparatus for automatic fake vehicle license plate discovery Disclosure Number: IPCOM000229576D
Publication Date: 2013-Aug-06
Document File: 3 page(s) / 89K

Publishing Venue

The Prior Art Database


We propose a principled method to make the identical car discovery in two images simultaneously and dependent on each other rather than sequentially and independently. The fake license plate issue has been drawing city government's attention for a long time. this disclosure proposes a principled and novel method and system for fake license plate discovery across multiple cameras. This disclosure discloses a method for simultaneously discover the identical car from two image samples, instead of the conventional methods that find the car in one image and find the corrsponding in the other image. By case study, we show that the proposed method can achieve superior discovery rate over conventional schemes. We develop a principled method and system to attack the specific problem - fake license discovery by simultaneously and iteratively considering the similarity between image-model, and comparing image samples.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 52% of the total text.

Page 01 of 3

System and apparatus for automatic fake vehicle license plate discovery

The disclosure is based on the following observations:

- The fake licensed car has the same appearance with the real one.

- The existing methods first extract the features of car A in image A, and then try to test if there is a similar enough car B in image B. Note it is ad-hoc to decide which features shall be used in testing the similarity: in general different features shall be used in different conditions.

The key task is 1) find the car area in the image, then further active tracking & refocusing can be made; 2) verify if cars are similar in two images; Both problems are challenging, can we perform them simultaneously to benefit to each other? - previous methods do it separately

The main idea consists of the two parts:

- Co-discovery from two images: consider the matching task in two images allows dynamically deciding which features shall be used in discovering similar cars in two images SIMULTANEOUSLY. It also avoids the difficult problem of plate number recognition which is even more challenging than identical car discovery.

- Re-focusing and tracking enabled by the car discovery: Once found the similar cars in two images, the further automatic similarity verification & plate recognition can be performed by an active vision system when the matching image's resolution is not ideal enough: adjust the camera to track the moving car in the reference image, or re-focus onto the car of interest, and re-focus on the plate area / driver face.

And additional ideas:

- Handle the 'escaping out of snapshot' problem by collaborative tracking: estimate the car moving direction in Camera A, and activate the most probable cameras where the car of interest will appear, and perform matching in these cameras to find the identical car and capture the higher-re...