Computer Vision (CV) involves techniques, including Deep Learning, that allow computers to “see” and process image data in a useful way. CV is of growing interest due to the explosion in image and video data being created and the rise of GPU-accelerated computing allowing its analysis. This vital discipline is making new kinds of analytics and experiences possible. Examples include:
- Identifying cancer in x-ray images
- Tagging friends and family in a personal photo library
- Scrutinizing video for evidence of lurking equipment defects
- Localizing objects out in the world using three-dimensional capture technology such as LiDAR
Primary Techniques
Image classification – determines whether an image satisfies a given criteria of interest. Examples include examining medical images for consistency with diagnoses and searching the frame of a video for a specific object.
Object detection and segmentation – identifies and locates multiple objects, of varying sizes, within a single image.
Generative Adversarial Networks – generates new, synthetic images by pitting two models against one another. One model generates images and another discerns whether a given image is real or synthetic. It is very useful – for many purposes — to be able to generate brand new images that plausibly could have come from your training dataset.
Computer vision techniques like these are also applicable to video data and three-dimensional data, with cutting-edge CV applications to autonomous vehicles and three-dimensional inputs such as LiDAR.
Applications
Image and video analysis and understanding – Anywhere image data is available, CV techniques can provide insights and support downstream analysis. Examples include:
Image classification
- Object detection and metadata generation
- Image deduplication and record linking
- Automatic captioning
Geospatial analytics and remote sensing – Whether examining two-dimensional satellite imagery or three-dimensional LiDAR data captured by a drone or land vehicle, computer vision techniques can extract useful higher-level constructs from raw image data. Examples include:
- Meteorology and climatology
- Augmented reality
Case Studies
- Iris: object detection
- Blackmarker: document redaction support
- Falcon: Lidar exploration
- Lottie: Objects, classification, facial recognition, image similarity,
- Geospatial: analysis of snow depth and type using satellite imagery