| Abstract |
Intelligent public transport is an integral component of smart cities. In this paper, a simpler approach has been demonstrated for autonomous taxi navigation. A single front camera with 1920x1080 pixel resolution at 60 frames per second has been used for real-time object detection and path planning. The deep neural network has been fed with the driving input of a couple of hours. The duly trained network was then tested with 30 minutes of driving data, recorded and used separately. Two control parameters were derived viz, speed of the vehicle and the steering angle. An eight layered convolutional network with two gated recurrent and four dense layers was trained and tested on GPU based system. The factors like speed, inclination, mean squared error (MSE) were 1.79 and 2.69 percent, respectively, with 38 milli seconds of real-time response delay. |
