Loss Prevention: The Latest AI Project
Artificial intelligence (AI) might not seem like a retail-centric technology, but don’t rush to conclusions — AI systems can help prevent shoplifting and enable loss prevention. Thus far, there’s been at least one AI technique investigated for dealing with this challenge: Scanning retail stores’ aisles via video gives a series of time-correlated images, which are then post-processed by AI to determine if the footage will show that a merchandise loss has occurred. If so, the system automatically generates an alarm.
But the trick to efficient security is enforcing reliable AI training, in accordance with the best practices of most current machine learning environments. Without ample (and valid) training data, it’s challenging to generate the rules that an AI will use to evaluate the probability of various outcomes.
Shopping and Copping: AI as Loss Prevention
One open-sourced effort to level up this technology is ShopoCop. The company has a partial set of software modules on Github that they’ve developed to take a retailer’s video feed and analyze it to spot shoplifters in real time.
For ShopoCop, theft detection by a video camera consists of two phases. The first is detecting and tracking the moving objects found in the video. Then, the system passes those objects to a frame-analysis model that classifies the motion. These models take the frame sequence and perform a number of transformations, which then get fed to the neural network. Convolutions and long short-term memory (LSTM) cells are used for analysis; ShopoCop says they use these because of their ability to detect complex visual patterns and memorize past events.
ShopoCop seems forward-thinking in their tone about the software, saying, “We will always keep the source code of the service and the models free and open for any retailer to use on-premise (in store). We believe there should be zero investment in software for any store that wants to protect its merchandise.”
Increased AI Use
The use of AI and video analysis in other areas of retail has been on the rise, too. As Fortune reports, Apple recently acquired Emotient, a company with a technology that can predict emotions from facial expressions on images. This system can help retailers not only prevent shoplifting but also better understand what a given shopper thinks of a particular merchandise display, so they can market to that individual more effectively.
If AI can scan store aisles to understand shoppers’ emotions, facial recognition as a form of customer identification might not be far behind. There’s certainly plenty of training data to utilize: Many stores already routinely collect video data of their retail area via DVRs, so no significant hardware investment is needed.
AI and Consumer Privacy
But this application of AI raises some important questions. What if a store’s loyalty program didn’t require a card scan at purchase time, because it already identified you by your facial features? How will that ID data be stored, and for how long? Would customers accept such an ID process? Ultimately, do these new technologies breach personal privacy and consent?
There’s a lot more than loss prevention involved when you mix AI with consumer video. While machine learning may be able to heighten security in certain situations, it isn’t always possible to bring the same level of security in the back-end environment to the store aisles. There are people involved here, not just networks — so while AI will certainly take on a greater role in retail, implementing it for loss prevention and facial recognition for identification may prove the biggest challenge.