Object Detection and Classification Algorithms using Deep Learning for Video Surveillance Applications

11/02/2022
by   Mohana, et al.
0

Object Classification is a principle task in image and video processing. It is exercised over a multitude of applications ranging from test and number classification to traffic surveillance. The primitive machine learning concepts had provided the pedestal for carrying out umber of image processing tasks. Nowadays requirement of detection algorithm is to work end to end and take less time to compute. Real-time detection and classification of objects from video provide the foundation for generating many kinds of analytical aspects such as the amount of traffic in a particular area over the years or the total population in an area. In practice, the task usually encounters slow processing of classification and detection or the occurrence of erroneous detection due to the incorporation of small and lightweight datasets. To overcome these issues, YOLO (You Only Look Once) based detection and classification approach (YOLOv2) for improving the computation and processing speed and at the same time efficiently identify the objects in the video. Classifier such as Haar cascade which uses Haar like features was primitively used for face detection. Moreover, due to the ever-increasing demand and scope of improvement in the existing fields, the primitive methods need a lot of upgradation. Neural Networks have made the tasks quite plain sailing. Right from the vanilla neural networks to Fast R-CNN and then Faster R-CNN, all models have contributed significantly in the domain of computer vision. This paper mainly focuses in detection and classification ranging from single class objects to multi class objects. The classification algorithm creates a bounding box for every class of objects for which it is trained, and generates an annotation describing the particular class of object. The Haar cascade classifier was trained on a batch of positive and negative samples which were later stitched together to form a vector file and finally form the xml file. On the other hand, COCO dataset used for implementing YOLOv2 and R-CNN algorithm due to the presence of pertained model in it. In addition, use of GPU (Graphics Processing Unit) to increase the computation speed and processes at 40 frames per second.

READ FULL TEXT

Please sign up or login with your details

Forgot password? Click here to reset