Breakthrough AI algorithms in computer vision
Our software incorporates breakthrough Artificial Intelligence and Machine Learning algorithms to address core industry problems of low recognition accuracy, large computational footprint, depth camera dependence, and massive training data collection. It has been designed to enable sophisticated contactless interaction at a fraction of the cost of conventional solutions.
21 joints skeleton with real-time positional coordinates
As opposed to traditional image-based approaches that treat hand as a monolithic entity, our software utilizes a 21 joints-based skeleton for each hand in camera view. Positional coordinates for all joints are provided in real-time for precise hand tracking. Skeletal approach can be extended to body tracking for assigning hand movements to individuals in a group.
100% hand tracking and gesture recognition accuracy
We deliver nearly 100% hand tracking and gesture recognition accuracy in any situation. Our software can track hand movements of any complexity from any camera angle. It can also recognize any static or dynamic gesture regardless of natural variations in its execution. Finally, hand tracking and gesture recognition can be performed with gloves and through glass.
10x reduction in custom gesture development costs
We do not require any training data collection for developing custom gestures upon request. Our AI/ML algorithms can generate joints-skeleton based training data synthetically and then extrapolate massively from a small amount of data. As a result, custom gesture development time and costs are reduced by a factor of 10 compared to conventional approaches.
Recognizes any static, dynamic, or drawing gesture
Our software can detect any type of hand gesture subject to camera specifications such as type, resolution, and lens. The evaluation SDK comes with commonly used static and dynamic gestures, including natural variations of these gestures. Custom gestures can be added to the SDK upon request at a fraction of development time and cost of conventional approaches.
Joints skeleton imposition on overlapping hands
We support tracking of overlapped hands using joints-skeleton approach regardless of camera position: front-facing or ego-centric. After overlapping, the system retains previously assigned IDs of individual hands for consistent identification in camera view. Positional coordinates for all joints in overlapped hand skeletons are provided in real-time with nearly 100% accuracy.
Joints skeleton tracking of occluded hands
We support tracking of occluded (i.e. hidden) hands using joints-skeleton approach regardless of camera position: front-facing or ego-centric. The system identifies occluded joints in addition to constructing joints-based skeleton for occluded hand portion. Positional coordinates for all joints in occluded hand skeletons are provided in real-time with nearly 100% accuracy.
Estimation of depth from 2D cameras
Our software can estimate depth (i.e. “z”) from ordinary 2D cameras that provide only “x” and “y” coordination. Depth information in turn allows construction of 3D joints-based skeleton for each hand in camera view. It also enables estimation of where index finger is pointing – for example, in space or on a digital screen - without conventional need to move one’s arm.
Assignment of hand movements to a person
We can track hand movements of individuals in a group using joints skeleton approach. Body skeleton covers 13 key joints in upper body such as eyes, ears, nose, shoulders, elbows, wrists, and palm center. It is linked to a 21 joints skeleton for each hand attached to the body, thereby, allowing hand movements and gestures to be assigned to a specific person in a group.
Operable with RGB, NIR, and/or Depth cameras
Our hand tracking and gesture recognition software is camera agnostic. It can be used with any 3rd party RGB (i.e color, monochrome), NIR (i.e. near infrared), and Depth (e.g. time of flight, stereo vision) camera. We can also customize SDK for desired camera upon request by addressing parameters such as imager resolution, field of view, and camera position.
Deployable on ordinary microprocessors
Our SDK is computationally efficient and can be deployed on ordinary processors without requiring hardware acceleration. For instance, for average webcam cameras, it can support hand tracking and gesture recognition with a single hand on ARM Cortex A53 (or equivalent). The number of supported hands can scale according to available computational resources.
Application agnostic software for any sector
We can support any use-case for contactless interaction with any product, in any industry. Some of the sectors in which we are pursuing commercial engagements with OEMs, include: automotive, banking, consumer appliances, digital signage, healthcare, home automation, hospitality, mobile phones, metaverse, retail, robotics, toys, and video games.
