Object Detection Engine

CLIENT

Client is a one of premier firms in USA, providing IT services and solutions for transportation industry.

OBJECTIVE

To process sequences of images for presence or absence of seal(s) behind the container for use by authorized marine terminal operators.

PROJECT SCOPE

To design and develop the camera based Object  Detection Engine. The engine was going to be integrated as an add-on plug-in with existing client’s system. The existing system reads video frames from video pools, and matches them against object templates—the results of which are stored in a central database. The central database maintains one record for each job. Jobs (and their associated images and job ticket) are stored in one transaction folder on the hard disk. Within each job folder is a job ticket—a text file containing key data elements such as transaction ID, site name, camera name, date/time stamp, and image location.

KRITIKAL'S ROLE

After understanding the existing system at Client’s end, KritiKal designed and developed an independent module for Object Detection Engine (ODE). The software for ODE consists of an Object Detection Interface (ODI) module that encompasses a Localization module (LM), Detection Module (DM) and a Correlation Module (CM).

At initialization, the existing system initiates an object of the ODI module and passes on the input parameters and the name of a configuration file to it. These parameters control the behavior of the different functional components viz. localization of seal in images, detection of seal in localized images and consolidation of individual results to derive final result for the transaction. The ODI reads in the configuration file and initializes itself and its sub-modules. This initialized ODI object is then used by existing system to initiate process of seal detection for transactions.

Once the process of seal detection is initiated by existing system, ODI first invokes the Localization module to identify probable regions containing seals in each image of transaction. Once the localization module has done its job, ODI feeds all marked images to the Object Detection Module (ODM). ODM comes up with an affirmative and negative response on every image for presence of seal with certain accuracy. After this, ODI would collate all these individual results from the individual images and pass it on to Consolidation Module (CM) to generate a consolidated result of seal detection.

ODI then communicates the final result to existing system as return value and along with confidence level.