Computer vision is a broad science that contains the objects detection and recognition as Important aspects. In this section, 1st we review SSD, fast R-CNN, faster R-CNN, mask R-CNN and the different versions of yolo model that used to detect object and give their advantages and disadvantages. 2nd we review lbph algorithm with haar cascade that used in face recognition. object detection and classification in real time is still a challenging job.
There are many traditional methods used in the object identification and classification such as HOG feature and SVM classifier. But it was expensive and slow. regional convolution neural network (R-CNN) solved this problem. It used selective search method which reduced number of bounding boxes that are fed to classifier to 2000 region proposals[12]. but it is quite slow .
One reason is that it requires a forward pass of the CNN for every single proposed region for every single image[13]. Fast regional convolution neural network (Fast R-CNN) is the next version of the R-CNN. Fast R-CNN consumes less time and improves the accuracy in detection. In Fast R-CNN, Feature extraction is done on the image before generating the object proposals. this will be enough running one CNN over the entire image. it Using softmax classifier instead of SVM. Then for every object, a pooling layer called Region of Interest (RoI) will extract a fixed-length feature vector from the feature map. Thus, obtained feature vectors are fed onto a series of fully connected layers. For this reason the model been fast than R-CNN, end to end trainable and higher detection accuracy. Fast R-CNN is still not enough speed to detect in real-time. Both R-CNN and Fast R-CNN uses selective search to find out the region proposals.
Selective search is a slow and time consuming process affecting the performance of the network. Faster regional convolution neural network (Faster R-CNN) is version 3 of R-CNN. Faster R-CNN adds a fully-convolutional network on top of the features of the CNN known as the RPN instead of selective search to speed up the region proposal. Faster R-CNN is much faster than version above. Thus, it can even be used for real-time object detection. Mask R-CNN is an extended version of faster R-CNN. It works by inserting a branch into faster R-CNN that adds a binary mask to determine whether a given pixel is part of an object. Single-Shot Detector (SSD) is deep neural network model for performing object detection. This model is having the capacity to combine the predictions generated from the different feature maps with a variable resolution to handle objects of distinct size. Single shot detectors are faster but lack in accuracy.
In 2016 Joseph Redmon and his colleagues present yolo model. Firest version So this article uses YOLO V3 as the insulator defect detection algorithm[].in face recognition we have implemented LBPH algorithm with Haar cascade which used in real time. LBPH has been used for face recognition. Haar cascade has been used for facial identification.
References
- Computer Vision: Object Detection beyond Object Detection
- SSD: Single Shot MultiBox Detector
- Fast R-CNN
- Faster R-CNN
- Mask R-CNN
- YOLO: Real-Time Object Detection
- Eye tracking for detection and classification of fatigue development in drivers
- Towards Machine Learning on Blogs – Web Scraping with Python