<p>Accurate and timely accident detection is crucial for enhancing road safety and emergency response systems. In this study, we propose a CNN+LSTM-based deep learning model for accident detection using sequential image data, aiming to leverage spatial and temporal features for higher predictive accuracy. We evaluated the efficiency of the proposed CNN+LSTM model by comparing its performance with that of CNN+GRU and Simple CNN.According to the experimental data, the CNN+LSTM model performs better in terms of accuracy and loss than both CNN+GRU and Simple CNN. The proposed CNN+LSTM model achieves a training accuracy of 90.22%, validation accuracy of 91.30%, and testing accuracy of 91.31%, with a corresponding training loss of 0.2113, validation loss of 0.2306, and testing loss of 0.1840. In comparison, the CNN+GRU model attains a training accuracy of 87.55%, validation accuracy of 86.96%, and testing accuracy of 87.62%, with a training loss of 0.2865, validation loss of 0.3371, and testing loss of 0.2597. The Simple CNN model records a training accuracy of 89.30%, validation accuracy of 90.82%, and testing accuracy of 90.90%, with a training loss of 0.2769, validation loss of 0.2592, and testing loss of 0.2561.</p>