In the next few years, our attention, behavior, and emotions will be tracked in real time by a variety of devices and systems including smartphones, webcams, security systems, and possibly cars. The images and videos we post online will be analyzed increasingly by automated systems, extracting data about gender, age, health, and personality, among a growing list of other traits. The impetus for these changes primarily comes from advances in computer vision and artificial intelligence, enabling new research possibilities and commercial applications.
Unlike traditional approaches to tracking eye movements and behavior – which are predominantly focused on hardware and confined to laboratory settings – new online, software-based systems are platform agnostic, adaptable, and can analyze groups of individuals in real-world settings. For example, using only video feeds or images, Sightcorp’s technology can track seven types of emotion, head position, head gaze, eye gaze, gender, ethnicity, mood, and clothing style, among other features. It scales from individuals to groups, and it can be used across multiple platforms, including mobile. A demo of the company’s software can be found here.
Eye tracking alone has been researched for more than 100 years, beginning in the late 19th century with crude and invasive mechanical systems that involved direct contact with the cornea. Although major advancements were made over time, today’s commercially available eye trackers still embody technological innovations of the 1960s and 1970s. Only recently have computational capabilities reached a point where software and algorithmic solutions became tenable.
As such, hardware-focused eye tracking companies such as Tobii, Applied Science Laboratories, and Grinbath are attempting to keep up with the changing technological landscape – even offering tetherless eye trackers and mobile extensions – but they are ultimately plagued by slow development cycles, costly equipment, and an inability to scale. For instance, Grinbath’s Mobile Tracker cost $5,500 per unit. Collecting data on just 10 individuals at once would require over $50,000 in equipment, not including training on proper use of the device, as well data processing and analysis.
Comparatively, the solutions of companies like Sightcorp and AlchemyAPI reduce the cost of crowd attention and behavior analysis by orders of magnitude without the need for bulky hardware or device compatibility issues. It will be interesting to see how traditional providers evolve to remain competitive, especially as the accuracy and validity of software-focused systems increases over time.
Some trepidation exists, however, with emerging behavior tracking systems. Privacy concerns will undoubtedly fuel public discourse over the intricacies of profit and anonymity. And potential dangers such as identity theft and victimization need to be addressed. Nonetheless, despite arguments against pervasiveness and scope, solutions based on AI and computer vision are increasingly advantageous over traditional approaches. Novel algorithmic developments and increases in processing power, for example, can have profound effects on performance and scaling capabilities, despite being low cost and relatively easy to instantiate. In addition, the cost performance of high-resolution video systems like CCTV cameras is dropping exponentially, falling from an average of $3,000-$5,000 in 2010 to $600-$800 as of 2015, and projected to be around $100 by 2020. This will further drive the ubiquity of these new systems.