Abstract
Pedestrian detection is at the core of autonomous road vehicle navigation systems as they allow a vehicle to understand where potential hazards lie in the surrounding area and enable it to act in such a way that avoids traffic-accidents, which may result in individuals being harmed. In this work, a review of the convolutional neural networks (CNN) to tackle pedestrian detection is presented. We further present models based on CNN and transfer learning. The CNN model with the VGG-16 architecture is further optimised using the transfer learning approach. This paper demonstrates that the use of image augmentation on training data can yield varying results. In addition, a pre-processing system that can be used to prepare 3D spatial data obtained via LiDAR sensors is proposed. This pre-processing system is able to identify candidate regions that can be put forward for classification, whether that be 3D classification or a combination of 2D and 3D classifications via sensor fusion. We proposed a number of models based on transfer learning and convolutional neural networks and achieved over 98% accuracy with the adaptive transfer learning model.
Highlights
Autonomous vehicles are becoming increasingly prevalent on roadways around the world; a study conducted in 2020 by Mordor Intelligence [1] reports that “the autonomous car market was valued at USD 20.97 billion in 2020” and is projected to increase by 22.75%, to USD 61.93 billion by 2026
Table presents the overview of ative samples were the CVC-02 data splits used in the development of pedestrian classification models
InIn this sub-section, we we describe three models developed via transfer learning: the first first two of which the convolutional blocks in order to preserve the original two of which “freezes” the convolutional blocks in order to preserve the original VGG-16
Summary
Autonomous vehicles are becoming increasingly prevalent on roadways around the world; a study conducted in 2020 by Mordor Intelligence [1] reports that “the autonomous (driverless) car market was valued at USD 20.97 billion in 2020” and is projected to increase by 22.75%, to USD 61.93 billion by 2026. Models trained using ImageNet may be exceptional at differentiating between a wide variety of classes, applying such a model to a more specific use-case would likely result in a significant loss of performance. As discussed in earlier sections, deep learning and machine learning models require significant volumes of data for use during training This was identified in 2001 in a research report published by Gartner [33], which alluded to the impending surge of big data. Transfer learning (TF) aims to provide a middle ground, where knowledge acquired from larger datasets can be used in conjunction with smaller, domain-specific, datasets in order to improve performance in subsequent domain-specific tasks In this context, prior knowledge can be model weights or low-level image features which describe what is being classified such as edges, shapes, corners, pixel intensity, etc.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have