A Multi-Target Detection Algorithm for Infrared Image based on Retinex and LeNet5 Neural Network

Abstract

Object detection in infrared video images is an important and challenging work. Due to low resolution, poor contrast, and low visual quality, target detection in infrared images is inefficient and prone to having higher false positive and lower precision rates. To improve detection efficiency, according to the characteristics of infrared images, we proposed a multi-target detection algorithm based on image enhancement and the LeNet5 deep neural network. In our method, we used the Retinex image enhance algorithm to protrude the edge contour and contrast, highlight the detailed features, and enhance the overall visibility of infrared images. In particular, the LeNet5 convolution neural network and CVC vehicle-assisted driving database were used to train the interesting target in the infrared image to generate the target data model, and the selective search algorithm was used to segment the candidate detect object regions in the image. The separated candidate regions were sent to the trained data model to classify the type and locate the position of objects in the image. The simulation results in CVC infrared image subset datasets show that our algorithm has higher detection speed and accuracy than the traditional HOG-based and LBP-based detection algorithms.