Cognitive Object Detection Assistant (CODA) - IBM Corporation
IBM won ACT-IAC’s “Transformer” award, at the Igniting Innovation 2018 Conference, for the innovation that uses existing technology to transform or extend existing capabilities resulting in new or broadened applicability and use.
IBM developed the Cognitive Object Detection Assistant (CODA) as a proof of concept. CODA proves early machine learning successes for X-ray baggage screening in the airport checkpoint environment and demonstrates the ability to extend and improve threat object detection performance in comparison to traditional detection methods. CODA augments the presented x-ray image for the operator by highlighting threat and prohibited items, identifying threat type, and providing a confidence score. CODA technology demonstrated a 99 percent detection rate for a handgun threat.
Checkpoint scanners are vital components to threat detection. Checkpoint scanners, built with an operator interface, present X-ray images of each bag to checkpoint operators. These operators have mere seconds to analyze any threats identified by a computer within the scanner and conduct a visual inspection of the bag for any prohibited items. This creates significant cognitive load on the operators as they visually search for nearly 100 prohibited items from gel candles to strike anywhere matches, while also resolving any computer-identified threats before the bag is declared safe to return to the passenger.
CODA provides an innovative approach at tackling two main deficiencies of the X-ray search process. Leveraging IBM Watson’s machine vision approached, CODA employs advanced machine learning techniques to rapidly identify threat. These techniques use models that are trained using supervised learning using tagged image data. At run time, CODA receives an image and compares it to all the threat object models. Selecting the best match, CODA tags the image with the object's name and confidence score. This resulting output is presented in an augmented view to the operator’s conducting the inspection.
Presently, operators must to visually inspect each bag regardless of the level of likelihood that a threat is contained. With further development, CODA could be used to pre-screen bags and only send bags with detected threats downstream for additional physical screening without the need to visually inspect bags through an intermediate process.
CODA provides a consistent and objective aid to facilitate standardized and improved operations by detecting and presenting objects of concern to the operator conducting checkpoint screening. CODA is able to tackle these challenges and potentially reduce the time required for operator analysis. CODA aids in operator threat detection in the checkpoint vetting process. In quantitative terms, CODA demonstrated early successes in detection of 99 percent of bags that had included handgun threat image projections. When trained, CODA can be used to detect any variety of objects. Qualitatively, CODA leads to better trust in checkpoint scanner results through high detection and low false alarm rates.