Low-Cost, Real Time Vehicle Classifcation and Speed Estimation using Raspberry pi 3 Vision Based and Ultra-Sonic Sensor Based Systems

This paper focusses on the design and development of a low-cost vehicle classification system using raspberry pi along with a speed estimation unit. This system, which is called augmented sensor-based classification system, uses a k-nearest neighbor (k-NN) method for the classification of vehicle classes based on vehicle dimensions. The vehicle class is strictly regulated by the law and is based on its dimensions. Hence with the aid of a camera and a few low-cost ultrasonic sensors we determine the speed and the dimensions of the vehicle. Experiments were conducted in laboratory with scaled models of different vehicle classes. The augmented sensor-based classification system has shown to classify vehicles with an accuracy of 0.842 with the speed estimation unit having an accuracy of 0.861. The system was tested for various camera angles and lighting conditions. We compared the performance of this system with a vision-based system using convolution neural network (CNN) trained directly on the images of vehicle belonging to different classes. The vision-based system had an accuracy of 0.858 which is highly dependent on the amount of training data, camera angle and lighting conditions. We see that performance of our low-cost system is comparable to that of the vision-based system.