Tired of skimming through paper after paper to figure out how to put your object detection model together? Alegion's data science team spends hours researching object detection techniques and localization methods, and which ones can be best applied to business applications.
We are putting these findings together and releasing a new series covering only the state-of-the-art concepts behind object detection R-CNN, Fast R-CNN, Faster R-CNN, and eventually Mask R-CNN.
Object detection is a sub-discipline of computer vision and image processing that detects certain classes of objects in digital images and videos. Popular applications of object detection include facial recognition software, object tracking such as a soccer ball during a game, or IDing pedestrians for autonomous vehicles.
Object classes are defined by a core set of properties without which a thing would not be what it is. A bicycle is a vehicle composed of two wheels held in a frame one in front of the other, propelled by pedals and steered with handlebars attached to the front wheel. If any of these core properties are changed, it is no longer considered a bicycle. For example, if another wheel is added, it becomes a tricycle. If a wheel is subtracted, it becomes a unicycle.
Object Detection methods fall into two main categories:
Computer Vision approaches:
- Viola–Jones object detection framework based on Haar features
- Scale-invariant feature transform (SIFT)
- Histogram of oriented gradients (HOG) features
Machine Learning approaches:
- Region Proposals (R-CNN, Fast R-CNN, and Faster R-CNN)
- Single Shot MultiBox Detector (SSD)
- You Only Look Once (YOLO)
Download a free copy of Part I of this series on R-CNN, created by our Applied Machine Learning Research Engineer, Saurabh Bagalkar, who specializes in Computer Vision.