Autonomous Path Finder and Object Detection using an Intelligent Edge Detection Approach
R Manasa, K Karibasappa, J Rajeshwari
Abstract
This work carried an approach to automatic detection and recognition of road boundary line demarcation as well as detection of other objects available over the road. Maintaining the vehicles over the road and in proper lanes is important for the autonomous vehicle and driver assistance system. The proposed work has applied an intelligent approach to edge detection using the feed-forward architecture of neural networks, which carried the self-adaptive strategy of transfer function slope in their active nodes. The proposed model has used the neural network as a low pass filter, which tries to develop the same outputs as available inputs. Other information from the image, except edges, was generated by the neural network very well. In contrast, the edges carry high-frequency information, and the neural network doesn't develop the output as it is. Hence a complement of the generated output image to the input image delivered the edges in the input image. The recognition of road boundary was detected as two parallel straight or curve lines. Detecting any object in the lane or over the road was done in two phases. In the first phase, the cover region was divided into several parts, and pixel density was estimated in the near region. The statistical information of pixels' distribution in the near and front region is estimated. In the second stage, a support vector machine-based classifier was used to define the object's presence. Such recognition can help control the vehicle's speed to run the vehicle safely. The proposed work has been applied to various real road images, and obtained results were appreciable. This work can be considered one step further toward developing autonomous vehicles cost-effectively.