A novel background updation algorithm using fuzzy c-means clustering for pedestrian detection

  • PDF / 3,448,888 Bytes
  • 15 Pages / 439.642 x 666.49 pts Page_size
  • 67 Downloads / 213 Views

DOWNLOAD

REPORT


A novel background updation algorithm using fuzzy c-means clustering for pedestrian detection Harshitha Malireddi1 · Kiran Parwani1 · B Rajitha1 Received: 3 May 2019 / Revised: 10 September 2020 / Accepted: 16 September 2020 / © Springer Science+Business Media, LLC, part of Springer Nature 2020

Abstract The task of pedestrian detection in video surveillance applications will face challenges like dynamic background changes, false human detection (shadow), and illumination variations. In literature, many approaches have been proposed to resolve these challenges. But their performance is not up to the mark. Thus this paper proposes efficient pedestrian detection including shadow removal and automatic dynamic background update. For this firstly, a background frame is initialized where no moving object is present. Then a background subtraction algorithm is applied to each of the key frames from the live video to detect the foreground objects (using fuzzy C means clustering followed by mean absolute difference). Later on this segmented foreground a contour is estimated and passed through the HOG classifier for pedestrian detection. The performance of the proposed approach has been compared using various datasets & state-of-the-art approaches and found to the best with an average precision of 98 %, unlike the others. Keywords Histogram of gradients · Precision · Fuzzy clustering · Membership matrix

1 Introduction Pedestrian identification has a wide range of applications in the current field of computer vision. It can be used for traffic and congestion analysis, human tracking, visual surveil Harshitha Malireddi

[email protected] Kiran Parwani [email protected] Rajitha [email protected] 1

Motilal Nehru National Institute of Technology Allahabad, Allahabad, India

Multimedia Tools and Applications

lance, fall detection, gender classification, robotics, human-computer interactions, etc. Over some time, human detection has turned out to be a popular research area having particular characteristics and features. In addition to this, human detection comes with its own set of challenges which include variations in illumination, pose, shadows, occlusions, and style. In the case of outdoor scenes or conditions that include rapid background changing scenarios have great impact on detection results. The complete procedure of finding individuals from images or videos can be summarized into three following steps: – – –

Identifying the region of interest i.e. Motion detection. Computation of effective features. Object classification i.e. identifying whether it’s human or not.

The most rudimentary way of identifying the object motion is to compare each frame with its previous frame, better known as background subtraction. Background subtraction [1, 3, 4, 12, 16, 17] is the most prevalent and robust method for detecting moving objects from static cameras by generating foreground masks. The process of modeling background involves steps related to initialization and updating the model to learn and adapt to v