| Summary: | With the high development of artificial intelligence, machine learning and pattern recognition are playing an increasingly important role in object detection and recognition from images or video sequences. This technique can help people to analyze and extract significant information of the image or video more efficiently and accurately. This dissertation conducts an experiment on vehicle detections from campus surveillance video sequences with Deep Learning Method using different datasets to pre-trained the network and compare the experimental results of each configuration. At first, we have searched and read materials about the development and state-of -art methods. After that with the help of Caffe framework and Python, we trained Neural Network with different types of datasets with different parameters and designed a Python program with Graphical User Interface (GUI) to show the results of detection. In the end, we compare the performance of detection results and find out that a fine-tuned pre-trained Neural Network can contribute to the improvement of detection performance.
|
|---|