Computer Vision – Enabling Enterprise Digital Transformation
Anik Bose (BGV General Partner) and Niranjan Venkatesan (BGV intern and Kellogg MBA student) share their perspective on innovation and value creation in computer vision. Computer vision is defined as tasks that include methods for acquiring, processing, analyzing and understanding digital images, and extraction of high-dimensional data from the real world in order to produce numerical or symbolic information, e.g., in the forms of decisions. In layman’s terms, computer vision is an arm of artificial intelligence (A.I.) that focuses on enabling machines to “understand” images by processing and analyzing them on a pixel-by-pixel basis, rather than relying on human-controlled categorization data, such as keywords and descriptions. Getting computers to recognize objects just like human beings do remains elusive; 100% accuracy and reliability is almost impossible. As a reference point, the human eye delivers 91% recognition. New technologies such as deep learning are evolving that promise to increase accuracy and reliability dramatically, but these technologies need more research before they can become mainstream. These technologies are modeled on a neural network and will allow the conversion of an image to text or speech. Deep learning will also enable the conversion of text or speech to image or video and the two together can enable tagging, searching, and indexing of a video or image just like text. Some of these new software technologies can deliver a recognition accuracy of 95%. Intellivision, a BGV portfolio company, is able to deliver 98% accuracy for specific vertical applications. Early M&A activity in the computer vision space by larger companies is a lead indicator of the large market potential. The most prominent acquisition being Intel’s recent $15bn acquisition of Mobile eye. Adoption Challenges While the computer vision market forecasts are significant, we believe there are several adoption challenges that must be addressed for the market opportunity to be realized. Key adoption factors include image data privacy concerns, lack of data accuracy, quality and ubiquity, automotive regulations, evolving industry standards and dependence on analytic platforms and applications. These factors, along with operational challenges such as costs, increases in video/image data driven by higher security needs, inability to maintain and fit legacy systems to the new requirements and the difficulties associated with scaling pilots in addition to high bandwidth requirements for hardware, are all rate limiting factors influencing the timing and size of the actual market opportunity. The Venture Investment Opportunity We believe that the computer vision technology stack can be broken down into three layers: a) hardware – This includes cameras, sensors, chipsets and video cards; b) solution frameworks – This includes image-processing algorithms, analytics/deep learning algorithms; c) application – This includes facial recognition, AR/VR and 3D imaging applications. This stack must be applied to proprietary data needed to train the algorithms required to solve specific customer problems. Access to such data will be the long-term differentiator for the winning companies, sometimes more important than the technology stack itself. We believe that there are six specific areas where innovative startups can play a strong role in value creation. These are:
- Robotics Vision – Driven by demands for the following: safety, quality, reliability, and ease of use, cost efficiencies, benefits of 3D over 2D technology, rapid deployment, and penetration of smart camera in retrofit robotic systems. Key early adopter industries will be automotive and food processing. 6dbytes is an innovative robotics software orchestration company that is targeting the food-processing industry.
- Visual inspection/Machine Vision – Manufacturers prefer machine vision for visual inspections for their high speed and repeatability of measurements. This accuracy, along with the ability to safely and reliably identify flows and defects in products without disrupting or delaying processes, will create strong demand for computer vision. Key early adopter segments are likely to be the electronics and automotive industries. Drishti is another innovative startup that aims to revolutionize the manufacturing assembly line by digitizing human actions based on technology that can enable non-intrusive and real time observation.
- Augmented Reality – Applications such as improved logistics in a warehouse, remote monitoring, trouble-shooting and hands free access to contextual information are expected to drive the adoption of AR. Retail and transportation are expected to be early adopter segments. Augment is a startup seeking to transform the retail shopping experience with AR. They are one of the first companies targeting the enterprise B2B space with an end- to-end AR platform.
- Video Analytics – Intelligent video surveillance to respond to rising threats along with increased demand for business intelligence and adoption of cloud and analytics are also creating the need for computer vision technology. Early adopter markets are likely to be public safety and defense. Intellivision is an innovative BGV portfolio company that provides intelligent video analytics and smart camera solutions for retail, smart home and IOT markets.
- 3D Imaging – The automotive and construction sectors are progressively adopting 3D imaging solutions for designing, previewing and rectifying the final version of the end products. Fast-paced adoption is expected in healthcare as well for surgical applications and diagnosis.
- Image/Facial Recognition – Applications will increasingly value code recognition, digital image processing, object recognition, pattern recognition and optical character recognition. Media and entertainment is expected to be a lead early adopter segment. Finally, facial recognition will be driven by applications such as law enforcement, surveillance, mobile device authentication as well as targeted advertising in retail.