Vehicle and Pedestrian Detection in Urban Traffic Based on SSD-DeepSort

With the acceleration of urbanization and the continuous increase of traffic flow, using deep learning technology to enhance the accuracy and efficiency of vehicle and pedestrian detection systems has become a practical and effective solution. To address the issues of low detection accuracy and sensitivity to environmental factors in traditional sensor based detection systems. This paper proposes a pedestrian and vehicle detection method for urban roads based on SSD (Single shot MultiBox Detector) and DeepSort model. Firstly, obtain the original image based on urban road videos and perform image preprocessing. Then, based on the SSD model, prior boxes are generated through different feature layers to detect targets of different sizes, improving the detection speed and accuracy of the model. Finally, based on the DeepSort model, the tracking accuracy and robustness were improved by introducing object appearance feature extraction and similarity calculation. The experimental results show that the SSD algorithm has significant advantages in detection speed and accuracy, while the DeepSort algorithm effectively handles target occlusion and intersection, achieving high-precision tracking. However, there are still some errors and missed detections when dealing with small targets or complex scenes.