PROJECT TITLE :

From Handcrafted to Deep Features for Pedestrian Detection A Survey

ABSTRACT:

Detecting pedestrians is an important but difficult problem in the field of computer vision, particularly in activities that are focused on people. With the assistance of handcrafted features and deep features, a considerable amount of progress has been seen over the course of the past decade. The following is an exhaustive review of recent developments in pedestrian detection that have taken place. In the first step of this process, we offer a comprehensive analysis of single-spectral pedestrian detection, which covers both handcrafted features-based methods and deep features-based approaches. When it comes to methods that are based on handcrafted features, we present a comprehensive review of different approaches and discover that handcrafted features that have a large number of degrees of freedom in both shape and space have superior performance. In the case of approaches that are based on deep features, we divided them into two categories: those that only use CNN-based features and those that use features that are handcrafted in addition to CNN-based features. We provide a statistical analysis and trend of these methods, noting that feature-enhanced, part-aware, and post-processing methods have garnered the majority of the attention. In this article, we not only discuss the detection of pedestrians using a single spectral band, but also multi-spectral pedestrian detection, which offers more robust features for varying levels of illumination. In addition to this, we present a comprehensive experimental analysis as well as some relevant datasets and evaluation metrics. As a way to wrap up this survey, we will focus on open issues that still need to be resolved and will highlight the many different possible future paths.