Comparative Study of Object Shape Recognition using Ultrasonic Sensor Arrays with Artificial Neural Network

In this paper, attempts are made to detect some basic shape of objects using ultrasonic sensor arrays with Artificial Neural Network (ANN). For this purpose two types of ultrasonic sensor array are designed on prototype basis using HCSR04 ultrasonic sensors. The time of flight (TOF) variations of the sensors are considered as the features for the recognition unit. The objects are kept at a fixed distance from the sensor array to measure the TOF. The TOF of different sensors varies with the different object's shape. The two tools of ANN named `Function Fitting' and `Pattern Recognition' are used as shape recognition unit. These tools are trained with particular objects data. During the testing phase, the previously trained neural networks are recalled to detect the desired shape of the objects. Based on the detection results of two ANN tools a comparative study is redacted between the hexagonal sensor array and 3x3 sensor array. The study observed that, the 3x3 sensor array under `Function Fitting' tool gives the better result of shape detection. Moreover, this system avoids the complexity of signal conditioning algorithm and circuitry. It can be incorporated within a smart device to aid the blind people if this system is fine-tuned properly.