Obstacle Detection Using Sensor Fusion and Deep Neural Network for Motion Control of Smart Electric Tractor

The need for obstacle detection is quintessential from the safety point of view for modern smart/autonomous vehicles. The implementation of such technology in farm equipment can lead to further improvements in efficient farming. This necessitates the requirement of a low cost and reliable method for obstacle detection and motion control. To suffice the need, this research work is focused on the development of a perception module using multiple sensors which can act harmoniously in a given scenario. To detect the obstacle, three different sensors are used, providing the distance and feature of the obstacle. The camera is used for object detection and distance measurement using OpenCV deep neural network. As the simultaneous distance measurement is relatively slow and dependent on the environmental conditions pertaining to visibility, a mini Lidar module is used. As the Lidar module has a limited field of view, ultrasonic sensors are used for the detection of obstacles at close range. Data obtained from the system is used to drive commands for the vehicle's motion using a set of actuators controlling the vehicle's motion in terms of acceleration, braking and steering.