<p>Natural and man-made disasters are a major danger to the lives of people involved in the rescue process, urban search and who are directly affected. With the evolution of deep learning and Artificial Intelligence, automatic victim detection systems are substituting rescue teams. These systems utilize crucial sensors like infrared cameras, visual imagers, etc., to extract the data. Infrared imaging enables detection in adverse weather conditions, even during the night. Artificial Intelligence leads to reduced life risk, impact to society, respond to disaster situation and damage to the environment accurately.&#xa0;In this research work, a novel model is designed that uses both the thermal and skin-color parameters to detect the live human bodies (victims) when catastrophes occur like earthquakes, tsunamis, etc. Skin detection techniques are applied to color images. An Improved Stacked Sparse Autoencoder (I-SSAE) technique is designed to mine the features. Thereafter, extracted features are passed to Deep Learning and Machine Learning classifiers to have an Automatic Detection System (ADS). Three local feature descriptors HOG (Histogram of Oriented Gradients), SIFT (Scale Invariant Feature Transform) and SURF (Speeded Up Robust Features) are obtained on infrared images to compare the proposed technique.&#xa0;The performance of three local feature descriptors and proposed I-SSAE have been considered in the perspective of human recognition. Furthermore, a new dataset is created, which is acquired by FLIR E60 infrared camera. An intensive study is performed using several deep learning and machine learning techniques. In comparison to deep learning and machine learning models, I-SSAE + BiLSTM hybrid deep learning model is the most accurate combination to predict the victim’s body. The suggested model is performing well due to their ability to learn temporal and spatial information from the input data</p>