SimpleNet: Hazardous-Object Detecting Neural Network in Low-Performance Devices

While the widespread of mobile devices has brought convenience in many aspects, it has also caused an increase in the number of traffic accidents due to pedestrians preoccupied with their smartphone devices. In order to solve such problem, this research proposes a network that allows training of data with reduced number of layers. SimpleNet algorithm simplifies the process of object classification into the states of detected or undetected. In order to optimize the configuration for a given device, two hyperparameters, Repeat Parameter and Feature Parameter, were adjusted throughout the testing. The test results show that the performance of full- size SimpleNet is comparable to that of Deep Neural Network. Even after reducing the model with optimized hyperparameters, SimpleNet exhibited an accuracy of 96.8%. We expect that this algorithm can be implemented in low-performance devices to detect hazardous objects on streets. Furthermore, we expect the detection target to be more limited for these devices in that they can only have a restricted vision of the surroundings.