• Collecting training imagery dataset
• Choosing an appropriate algorithm
• Image quality control, grouping and labelling the images
• Implementing and training the CNN model
• Evaluating initial results
• Data augmentation
• Model fine-tuning
• Applying the trained model to crosswalk detection in Delft
• Testing model with different spatial resolution imagery

The model exhibited a tendency to misidentify non-crosswalk linear objects as crosswalks, resulting in many false positive predictions. This issue was found to be caused by unclear training data. Data had been included which not only explicitly depicted roads, but also included wider urban and sub-urban regions where objects like windows and solar panels caused the model’s confusion. This can be solved by filtering the predicted crosswalk locations using a road buffer (which was done as part of this research), by pre-processing the detection dataset to filter out non-road areas or by including appropriate images as background true negative samples to the training dataset. 

The trained model was also tested with imagery of higher spatial resolution and image quality than the ones used to train it. While some additional crosswalk locations were identified this way, predictions on the high-resolution image included a dramatically increased number of false positives. This was attributed to the sharper depiction of background linear objects due to the higher image quality and spatial resolution, making them more susceptible to misidentification. Training a new model with high resolution imagery with an appropriately sized true negative sample would possibly help reduce the number of erroneous predictions. Such an approach would however greatly increase computational demands and make the designed system unsuitable for real-time detection.