feat(multi_object_tracker): add object class filtering in tracking pr…

…ocess (autowarefoundation#6607) * feat: object class filter Signed-off-by: Taekjin LEE <[email protected]> * fix: set a member private Signed-off-by: Taekjin LEE <[email protected]> * fix: last filtered label is not useful, remove Signed-off-by: Taekjin LEE <[email protected]> * style(pre-commit): autofix Signed-off-by: Taekjin LEE <[email protected]> * fix: multiply gain for new class Signed-off-by: Taekjin LEE <[email protected]> * style(pre-commit): autofix Signed-off-by: Taekjin LEE <[email protected]> * chore: algorithm explanation Signed-off-by: Taekjin LEE <[email protected]> * fix: revise the filtering process flow Signed-off-by: Taekjin LEE <[email protected]> --------- Signed-off-by: Taekjin LEE <[email protected]> Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
tier4 · Mar 25, 2024 · f10a592 · f10a592
1 parent f6baf4b
commit f10a592
Show file tree

Hide file tree

Showing 4 changed files with 65 additions and 2 deletions.
diff --git a/perception/multi_object_tracker/include/multi_object_tracker/tracker/model/tracker_base.hpp b/perception/multi_object_tracker/include/multi_object_tracker/tracker/model/tracker_base.hpp
@@ -42,6 +42,8 @@ class Tracker
   {
     classification_ = classification;
   }
+  void updateClassification(
+    const std::vector<autoware_auto_perception_msgs::msg::ObjectClassification> & classification);
 
 private:
   unique_identifier_msgs::msg::UUID uuid_;

diff --git a/perception/multi_object_tracker/src/tracker/model/multiple_vehicle_tracker.cpp b/perception/multi_object_tracker/src/tracker/model/multiple_vehicle_tracker.cpp
@@ -43,7 +43,7 @@ bool MultipleVehicleTracker::measure(
   big_vehicle_tracker_.measure(object, time, self_transform);
   normal_vehicle_tracker_.measure(object, time, self_transform);
   if (object_recognition_utils::getHighestProbLabel(object.classification) != Label::UNKNOWN)
-    setClassification(object.classification);
+    updateClassification(object.classification);
   return true;
 }
 

diff --git a/perception/multi_object_tracker/src/tracker/model/pedestrian_and_bicycle_tracker.cpp b/perception/multi_object_tracker/src/tracker/model/pedestrian_and_bicycle_tracker.cpp
@@ -43,7 +43,7 @@ bool PedestrianAndBicycleTracker::measure(
   pedestrian_tracker_.measure(object, time, self_transform);
   bicycle_tracker_.measure(object, time, self_transform);
   if (object_recognition_utils::getHighestProbLabel(object.classification) != Label::UNKNOWN)
-    setClassification(object.classification);
+    updateClassification(object.classification);
   return true;
 }
 

diff --git a/perception/multi_object_tracker/src/tracker/model/tracker_base.cpp b/perception/multi_object_tracker/src/tracker/model/tracker_base.cpp
@@ -54,6 +54,67 @@ bool Tracker::updateWithoutMeasurement()
   return true;
 }
 
+void Tracker::updateClassification(
+  const std::vector<autoware_auto_perception_msgs::msg::ObjectClassification> & classification)
+{
+  // classification algorithm:
+  // 0. Normalize the input classification
+  // 1-1. Update the matched classification probability with a gain (ratio of 0.05)
+  // 1-2. If the label is not found, add it to the classification list
+  // 2. Remove the class with probability < remove_threshold (0.001)
+  // 3. Normalize tracking classification
+
+  // Parameters
+  // if the remove_threshold is too high (compare to the gain), the classification will be removed
+  // immediately
+  const double gain = 0.05;
+  constexpr double remove_threshold = 0.001;
+
+  // Normalization function
+  auto normalizeProbabilities =
+    [](std::vector<autoware_auto_perception_msgs::msg::ObjectClassification> & classification) {
+      double sum = 0.0;
+      for (const auto & class_ : classification) {
+        sum += class_.probability;
+      }
+      for (auto & class_ : classification) {
+        class_.probability /= sum;
+      }
+    };
+
+  // Normalize the input
+  auto classification_input = classification;
+  normalizeProbabilities(classification_input);
+
+  // Update the matched classification probability with a gain
+  for (const auto & new_class : classification_input) {
+    bool found = false;
+    for (auto & old_class : classification_) {
+      if (new_class.label == old_class.label) {
+        old_class.probability += new_class.probability * gain;
+        found = true;
+        break;
+      }
+    }
+    // If the label is not found, add it to the classification list
+    if (!found) {
+      auto adding_class = new_class;
+      adding_class.probability *= gain;
+      classification_.push_back(adding_class);
+    }
+  }
+
+  // If the probability is less than the threshold, remove the class
+  classification_.erase(
+    std::remove_if(
+      classification_.begin(), classification_.end(),
+      [remove_threshold](const auto & class_) { return class_.probability < remove_threshold; }),
+    classification_.end());
+
+  // Normalize tracking classification
+  normalizeProbabilities(classification_);
+}
+
 geometry_msgs::msg::PoseWithCovariance Tracker::getPoseWithCovariance(
   const rclcpp::Time & time) const
 {