<p>Surveillance networks using Internet of Multimedia Things (IoMT) technology involve embedded video sensors capturing periodic video frames transmitted to the edge gateway and then to the cloud. The IoMT surveillance networks produce vast amounts of multimedia data, which presents a significant challenge in effectively managing and handling this extensive data within the IoMT network. The process of sending all captured images to the edge gateway and then to the cloud can lead to increased data transmission, high energy usage during transmission, augmented communication overhead, latency, and potential bottlenecks at the edge gateway level. To address these challenges, this paper proposes an energy-efficient video data reduction (EViDaR) for edge computing-based IoMT surveillance networks. EViDaR operates periodically across two levels: video sensor devices and the edge gateway. At the video sensor device, the EViDaR applies a transmission video reduction using the Orientated FAST and Rotated BRIEF algorithms (ORB) for intra-view summarization. A deep multi-view video Summarization will be applied at the edge gateway. It involves collecting data from different video sensor devices, processing the data, identifying shot boundaries, and generating a multi-view video summary enriched with features extracted from VGG16 (Visual Geometry Group). Finally, the summarized video is produced using a deep reinforcement learning-based LSTM neural network. We conduct several simulation experiments and compare them with existing data reduction and video summarization strategies, and the findings show the superiority of the proposed EViDaR in terms of frame rate adaptation, data reduction, transmitted frame count, precision, recall, and F-measure compared with other methods.</p>