Classifying compartments at security checkpoints by detecting a shape of an object

#The present invention relates to systems and methods for screening compartments for prohibited objects, and more specifically the embodiments of a classification system for automatically detecting prohibited objects in a compartment at a security checkpoint.

#Security checkpoints for screening compartments, such as luggage, backpacks, and suitcases are located at almost all transportation hubs. The screening process must be thorough yet also allow for a high volume of travelers to pass through the security checkpoint without sacrificing a quality of the screening process.

#An embodiment of the present invention relates to a method, and associated computer system and computer program product for automatically detecting prohibited objects in a compartment at a security checkpoint. A processor of a computing system receives a three-dimensional representation of a compartment from an imaging device connected to the computing system. A region within the compartment that has a highest likelihood of containing a prohibited object is identified, based on a voxel classification of the three-dimensional representation using a first trained neural network. Shapes of objects located within the identified region are classified using a second trained neural network to determine whether any of the shapes correspond to a shape of a prohibited object known by the second neural network

#FIG. 1 depicts a block diagram of a classification system, in accordance with embodiments of the present invention.

#FIG. 2 depicts a block diagram of a screening procedure at a security checkpoint, in accordance with embodiments of the present invention.

#FIG. 3 depicts a system overview of the training of a first stage neural network, in accordance with embodiments of the present invention.

#In brief overview, security checkpoint locations at transportation hubs are designed to screen for prohibited items, such as explosives, sharps, weapons, and the like. To be allowed beyond a security checkpoint, travelers must allow security officials to scan or otherwise examine personal items, canyon bags, and other luggage. X-ray based technologies have been used for this purpose at security checkpoints for several decades. Items of interest, such as prohibited items, can be difficult to detect within this environment due to a range of orientation, clutter, and density confusion in a traditional two-dimensional (2D) X-ray projection. Specifically, the problem of objects occluding each other is a limitation of 2D X-ray scanners, which makes detection (automatically or by human operators) particularly challenging. For example, x-ray scanners take two images of an inside of a bag, which are displayed on a monitor for a security officer to inspect and visually determine whether the bag contains any prohibited items. X-ray technology at security checkpoints is limited because x-ray cannot penetrate certain objects that are often carried by travelers; some objects can obscure other objects.

#To improve detection and passenger experience, embodiments of the present invention utilizes Computed Tomography (CT) technology instead of x-ray technology. CT technology can image a bag as a series of images packed together to form a three-dimensional rendering that can be viewed at multiple angles. Computer algorithms analyze the three-dimensional volume of data to automatically classify a bag has having or not having a prohibited item. Thus, instead of the security officer manually inspecting images on a monitor, which takes time and with varied results based on an experience level of the security officer, embodiments of the present invention automatically detect a presence of a prohibited object based on a shape of the object and classifies the material as a potential threat or no threat. Many prohibited objects have recognizable and consistent shapes to allow for classification based on known shapes. Confident and automatic detection of prohibited objects within a traveler's bag at a security checkpoint saves time for the traveler while at the same time increasing the quality of the screening process.

#Image classification within three-dimensional space is often less studied than its two-dimensional counterpart due to increased spatial complexity, memory allocation, and relative lack of easily accessible data in the three dimensional space. However, advances in 3D sensing technologies and mature 2D deep learning architectures have garnered interest in identifying techniques and algorithms that can leverage greater computing power in 3D volumes, Due to structural components with relatively minor deformation, shape classification is a straightforward extension from the two-dimensional space into three dimensions. Attempting to capture both high-resolution and low-resolution spatial relationships by segmenting and classifying individual voxels classifications is computationally expensive. To address this issue, the following classification uses shape max orthogonal projections with CNN networks.