DeepEyes platform
Computer vision in automibile can be used to obtains visual data both inside cabin and outside a vehicle. For inside cabin it can be used for driver's behavior monitoring. While outside vehicle installation tells how well driving was as well as utilized as object information to autonomus driving car.
CCTV is everywhere but mostly not well utilized. With integration of DeepEyes platform, user can automatically gathering physical information of traffic, customer journeies, buying pattern that never exists on traditional POS data collection. It is also possible to perform real-time security alert of black listed or weapon detection.
Customer profiling is a fundamental task for any marketing campaign setup. Corporates may collect these data from member's registration. But in practice, buyer and member can be different person. For example, a daughter buy his father a bottle of juice using her dad's member code. Hence, more effective DeepEyes' customer profiling is required.
Our physical world has 3 dimentions of width, height and depth. However, taking a picture cause dimention is reduced into 2 dimentions of width and height. The process of computer vision used for depth information recovery from these 2D images called '3D reconstruction'. The engine takes multiple 2D images as input and generate estimated 3D object as an output. Example applications including rapid 3D object prototyping, 3D map creation.
Image to text or "Optical Character Recognition: OCR" is an image processing technique that been used for long time. However, accurately recognize is a difficult task especially non latin characters. DeepEyes was designed to support scene text recognition as well as document text recognition using computer vision tasks.
The head pose estimation is a process of recovering 3D head position in term of yaw, pitch and roll from 2D images. In order to estimate head pose correctly, feature points tracking requires high precision. Our approach bases on the state space searching, the local minima problem is common. Hence, we divide the search space into sub spaces and perform parallel computation on GPU.