In the domain of surveillance and detection, acquiring distance information about targets is crucial. Traditional ranging methods, such as radar and laser ranging, have limitations due to their active emission of electromagnetic or laser signals towards the target during operation, which can potentially reveal their own position. To address this issue, an intelligent ranging algorithm based on monocular dual-modal image processing is proposed. This algorithm employs computer vision techniques and deep learning models to achieve non-contact passive ranging by capturing and analyzing images of the target. The process begins with obtaining a sequence of images of the target using a monocular camera, followed by feature extraction of these images through pre-trained deep neural network（DNN）models. It then combines principles of image formation with deep learning technology to estimate the relative distance between the target and the observation point by predicting ranging errors via neural networks. Under specific constraints, the ranging error of this algorithm is less than 10%. Ablation studies on continuous frames and long-term sequence correction strategies show that compared with single-frame input, the average error is reduced by 0.74 km, and the average percentage error is reduced by 3.69%, making it significant for applications in long-distance passive ranging. This approach represents a promising advancement in passive ranging technology, offering enhanced accuracy and reliability without compromising the concealment of the observation platform.