A Cognitive Approach To Predict the Multi-Directional Trajectory of Pedestrians
Jayachitra Virupakshipuram Panneerselvam Department of Computer Technology, Anna University, India This email address is being protected from spambots. You need JavaScript enabled to view it. |
Bharanidharan Subramaniam Department of Computer Technology, Anna University, India This email address is being protected from spambots. You need JavaScript enabled to view it. |
Mathangi Meenakshisundaram Department of Computer Technology, Anna University, India This email address is being protected from spambots. You need JavaScript enabled to view it. |
Abstract: Pedestrian detection is one of the important areas in computer vision. This work is about detecting the multi-directional pedestrian’s left, right, and the front movements. On recognizing the direction of movement, the system can be alerted depending on the environmental circumstances. Since multiple pedestrians moving in different directions may be present in a single image, Convolutional Neural Network (CNN) is not suitable for recognizing the multi-directional movement of the pedestrians. Moreover, the Faster R-CNN (FR-CNN) gives faster response output compared to other detection algorithms. In this work, a modified Faster Recurrent Convolutional Neural Network (MFR-CNN), a cognitive approach is proposed for detecting the direction of movement of the pedestrians and it can be deployed in real-time. A fine-tuning of the convolutional layers is performed to extract more information about the image contained in the feature map. The anchors used in the detection process are modified to focus the pedestrians present within a range, which is the major concern for such automated systems. The proposed model reduced the execution time and obtained an accuracy of 88%. The experimental evaluation indicates that the proposed novel model can outperform the other methods by tagging each pedestrian individually in the direction in which they move.
Keywords: Automated driving system, deep neural networks, faster recurrent-convolutional neural network, object recognition, pedestrian detection, pedestrian movement direction.
Received January 3, 2021; accepted March 20, 2022