With retail sales in their biggest slump since record-keeping began in 1992, businesses are desperately looking for ways to boost the bottom line. Improving productivity in existing stores with technologies like video analytics may turn out to be one of the most effective ways to accomplish this.
Sophisticated video analytics can help businesses assess shopping behavior by tracking customer behavior in aggregate and building up a repository of historical data. The information retailers are interested in has to do with where people are spending their time in the store, the effectiveness of store advertising displays, the demographics of visitors, and just generally how customer traffic is flowing through the store. The idea is to help retailers increase revenue and store productivity by optimizing merchandise choices and placement.
In the past, retailers convinced random customers to carry around tracking devices, so their shopping activities could be monitored. But this only provided a qualitative view of shopper behavior. Video analytics promises a quantitative view. For the most part, these analytics systems use video feeds from existing security cameras. But whereas the security video is mostly being examined manually — either in real time or as a result of some past mischief — video analytics can repurpose that same data stream to provide business intelligence for improving store operations.
Canadian startup LightHaus Logic is one of the pioneers in this area, providing high-performance video analytics applications for a variety of commercial settings. This week the company introduced Equinox Visual Intelligence solutions for the retail industry, which are currently under trial by a number of top retailers.
The LightHaus system distills terabytes of digitized video data into graphics-based reports that encapsulate customer activity over some a period of time. It aggregates a variety of information, such as the number of people at certain locations in the store and if they are standing still or moving (and in what direction). The system can also gather more specific information like effectiveness of the coveted end-of-aisle displays — the so-called “end-cap” displays. More sophisticated analytics, like identifying a shopper’s age, gender and ethnicity, can also be obtained in order to determine demographic shopping preferences.
Being able to condense hundreds or thousands of hours of video into actionable intelligence is the goal. “You don’t want to look at a month of video from several cameras,” explains Mario Palumbo, chief technology officer at LightHaus Logic. “You’d like a snapshot that shows you the whole story.”
The LightHaus offering uses a proprietary video gateway appliance and compute-intensive image processing software to convert the digitized video data into “metadata” that encapsulates customer activity in the store. The trick is processing multiple video feeds in real time. Because of the nature of the application, the hardware’s pedigree is from the embedded world, where real-time signal processing is a common type of workload. The LightHaus gateway appliance is a 1U box and uses multiple processors (general purpose CPUs and/or FPGAs) connected via PCIe or Ethernet links. An external disk is used for local storage of video and metadata.
Depending upon the resolution of the video feeds and the different types of analysis being run (occupancy numbers, demographics, display effectiveness, etc.), a single appliance will typically handle between five to 20 video streams. For large setups, multiple appliances can be hooked together. In this case the software auto-detects the other LightHaus boxes and one of them gets arbitrarily assigned to be the head node.
A certain amount of configuration is needed to set up system parameters like store layout and types of analysis. Palumbo says they try to make the configuration setup as easy as possible, since retailers typically don’t have a great deal of IT expertise. A lot of the larger operations like Home Depot and Wal-Mart employ system integrators to manage computer and video infrastructure, and these subcontractors would end up doing the configuration.
The most user-friendly part of the system is the report and data mining tools. The analytics-derived metadata is presented to the retailer in graphs and tables, with additional functionality provided for examining real-time and historical data. The whole system can be accessed via a standard Web browser interface, either locally or remotely. This allows large retailers to slice and dice the data produced from multiple stores and access it from any location.
LightHaus’ focus on business intelligence may prove to be a good move. Palumbo notes that although video analytics is a hot technology right now, most of the company’s potential competitors are focused on applying the technology to surveillance and security applications, where discrete event detection is the name of the game. In the retail world, businesses are less interested in event detection than in overall trends and aggregate statistics, which require a very different approach.
The current technology being offered can’t provide all the answers, though. For example, the LightHaus system is not able to determine which way the customer is facing. So in a store with different products in a narrow aisle, the analysis probably wont be able to determine what item caught the customer’s attention. Also, the cause of “hot spots” — where people linger for long periods of time — can be ambiguous. Hot spots might be the result of shoppers’ interest in a nearby product or their confusion by the related display.
But the systems are likely to become more sophisticated over time as hardware becomes more powerful and as providers like LightHaus refine the technology. At a time when retail shopping behavior is undergoing some big changes, video analytics may become a technology that businesses can’t afford to be without.