The multi-layer neural network applied to a car detection system
H. Nishihara, Ayaka Kojima, Hideki Murakoshi, Shintaro Ishijima · 2003
In image recognition, simple neural networks often recognize incorrectly due to the effect of changes in size and shifts in position of the object in the source image. The neocognitron is effective to recognize the image; however the network scale is too large, so it takes a long time to train the network and to process the information. The authors propose a car detection system using the simple multilayer neural network. This system has some pre-processors to improve the recognizing ability of the simple neural network. These pre-processors perform the role as a kind of regularization. This paper shows that the system effectively performs as a car safety system detecting other cars.>