In brief, YOLO Algorithm is an object detection algorithm. While YOLO stands for You Only Look Once. We use this algorithm to determine the class of an object. In other words, given an image, the algorithm determines the classes of objects in the image. Also, it determines the location of objects in the given image.
Hence, YOLO Algorithm is an important technique that we can use in many Computer Vision applications. In fact, the algorithm makes use of Deep Learning techniques. In order to do this, it applies a forward-only Convolution Neural Network for learning. Further, the algorithm takes a full image as input. Eventually, it predicts bounding boxes as well as class probabilities directly.
It is a real-time object detection algorithm. As can be seen, object detection is fundamental to self-driving vehicles. Because they need to monitor the objects coming in their way.
Why Image Classification is not sufficient?
Image classifications only tells us about which class an object belongs to. However, in certain computer vision applications like the ones used in self-driving vehicles we need more information. In fact, we need to know the position of the object also. So we need to determine where the object is there in the image. Usually, image classification only tells about what the image actually is. It doesn’t indicate the position of the object in the image. Moreover, image classification doesn’t work with images containing more than one object.
Basically YOLO is a CNN that identifies objects in real-time. At first, a single CNN is applied on the complete image. After that, the images are divided into regions. It predicts bounding boxes. Then it computes the probabilities for each region. Additionally, this method considers object detection as a regression problem.
This algorithm is referred to as “Look Once” since there is only one forward propagation pass through the CNN. Hence, it detects each object only once. The output of the algorithms comprises the detected objects along with bounding boxes. It is important to realize that YOLOalgorithm identifies both the object as well as its location in the image in a single algorithm run.
Features of YOLO Algorithm
Evidently, YOLO is very simple and fast. Also, it makes use of a single CNN. Another feature of YOLO is that it works in real-time. Further, it learns the generalized representation of objects. In other words, it can perform learning on natural images and identifies objects in an artwork. Comparatively, the number of background errors is lesser in YOLO. However, it doesn’t accurately detect smaller objects.
Applications of YOLO Algorithm
YOLO has applications in every situation where the goal is to identify objects in an image in real-time. Some of the examples are detecting traffic light signals, finding parking space, recognizing people from a photograph, autonomous vehicles, weed detection from the field image, plant disease detection, attendance monitoring, real-time surveillance, and so on.
Redmon, J., Divvala, S., Girshick, R., and Farhadi, A., 2016. You only look once: Unified, real-time object detection. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 779-788).
Huang, R., Pedoeem, J. and Chen, C., 2018, December. YOLO-LITE: a real-time object detection algorithm optimized for non-GPU computers. In 2018 IEEE International Conference on Big Data (Big Data) (pp. 2503-2510). IEEE.